{ "schema_version": "1.1.0", "generated_at": "2026-08-13T04:55:21Z", "price_catalog_generated_at": "2026-08-13T04:53:37Z", "suite": "MLPerf 6.0", "source_commits": { "inference": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "training": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421" }, "coverage": { "profiles": 12, "independent_submitters": 3, "independent_systems": 3, "matched_pricing_providers": 3 }, "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "price_context_count": 3, "price_contexts": [ { "matched_offer_id": "coreweave:b200:global:on-demand", "provider_id": "coreweave", "provider_name": "CoreWeave", "accelerator_model": "B200", "accelerator_variant": null, "region_scope": "global_or_unspecified", "regions": [], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 8.6, "per_gpu_hour_min": 8.6, "per_gpu_hour_max": 8.6, "required_gpus": 64, "cluster_hourly_rate": 550.4, "cluster_hourly_rate_min": 550.4, "cluster_hourly_rate_max": 550.4, "observed_at": "2026-08-13T04:53:37Z", "freshness_status": "recent", "source_url": "https://www.coreweave.com/pricing", "evidence_hash": "1d95e76dce83326c305bbf12abf1f36bb38a5bc1c4306dccbf3ebcdc1ebf9111", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized CoreWeave price is per GPU-hour; the benchmark system requires 64 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Region is not available in the current source.", "Live capacity is not available in the current source.", "CPU, RAM, storage and network inclusions are not normalized yet.", "The official source is an 8-GPU node; the displayed rate is normalized per GPU." ], "profile_ids": [ "coreweave-b200-64-gpt-oss-20b-training", "coreweave-b200-64-llama31-8b-training" ] }, { "matched_offer_id": "lambda:b200:global:on-demand", "provider_id": "lambda", "provider_name": "Lambda", "accelerator_model": "B200", "accelerator_variant": null, "region_scope": "global_or_unspecified", "regions": [], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 6.69, "per_gpu_hour_min": 6.69, "per_gpu_hour_max": 6.69, "required_gpus": 8, "cluster_hourly_rate": 53.52, "cluster_hourly_rate_min": 53.52, "cluster_hourly_rate_max": 53.52, "observed_at": "2026-08-13T04:53:37Z", "freshness_status": "recent", "source_url": "https://lambda.ai/pricing", "evidence_hash": "1621fbbdc19655ea81fda2416b5196b530d491dd82b5dc329a31b055b003f2ce", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Lambda price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Region is not available in the current source.", "Live capacity is not available in the current source.", "CPU, RAM, storage and network inclusions are not normalized yet." ], "profile_ids": [ "lambda-b200-8-gpt-oss-20b-training", "lambda-b200-8-llama31-8b-training" ] }, { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ], "profile_ids": [ "qwen3-vl-235b-a22b-offline", "qwen3-vl-235b-a22b-server", "deepseek-r1-offline", "deepseek-r1-server", "gpt-oss-120b-offline", "gpt-oss-120b-server", "gpt-oss-20b-training", "llama31-8b-training" ] } ], "profile_count": 12, "profiles": [ { "id": "qwen3-vl-235b-a22b-offline", "category": "batch", "benchmark": "qwen3-vl-235b-a22b", "scenario": "Offline", "dataset": null, "quality_target": "F1_HIERARCHICAL: 0.7880721729948121", "system": { "submitter": "Nebius", "platform": "nebius_b300_n1", "name": "Nebius B300 n1 (8x B300-SXM-270GB, TensorRT)", "status": "available", "division": "closed", "suite": "datacenter", "accelerator": "NVIDIA B300-SXM-270GB", "accelerator_count": 8, "software": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0" }, "measurement": { "metric": "throughput", "value": 78.2775, "unit": "Samples/s", "aggregation": "official_reported_result", "run_count": 1, "raw_values": [ 78.2775 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/inference-datacenter/", "rules_url": "https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc", "repository_url": "https://github.com/mlcommons/inference_results_v6.0", "details_url": "https://github.com/mlcommons/inference_results_v6.0/blob/4d3916ac9cf474b679cdfcf492d43a0559418ad1/summary_results.json", "submission_details_url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Nebius/results/nebius_b300_n1", "commit_sha": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "source_paths": [ "summary_results.json", "./closed/Nebius/results/nebius_b300_n1/qwen3-vl-235b-a22b/offline/performance/run_1" ], "content_hashes": { "summary_results.json": "52fa813a27834e8de38eca9fd381688df1cf9cd95020d2bd9e951f6cc16384ae" } }, "compatibility": { "provider_id": "nebius", "provider_name": "Nebius", "accelerator_models": [ "B300", "B300-SXM" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Nebius normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "current_reference_economics": { "kind": "throughput", "gpuCount": 8, "clusterHourlyRate": 62.8, "throughputPerSecond": 78.2775, "unit": "Samples/s", "referenceUnitCount": 1000, "costPerReferenceUnit": 0.222853878118801 } }, { "id": "qwen3-vl-235b-a22b-server", "category": "batch", "benchmark": "qwen3-vl-235b-a22b", "scenario": "Server", "dataset": null, "quality_target": "F1_HIERARCHICAL: 0.7865237675409595", "system": { "submitter": "Nebius", "platform": "nebius_b300_n1", "name": "Nebius B300 n1 (8x B300-SXM-270GB, TensorRT)", "status": "available", "division": "closed", "suite": "datacenter", "accelerator": "NVIDIA B300-SXM-270GB", "accelerator_count": 8, "software": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0" }, "measurement": { "metric": "throughput", "value": 45.1513, "unit": "Queries/s", "aggregation": "official_reported_result", "run_count": 1, "raw_values": [ 45.1513 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/inference-datacenter/", "rules_url": "https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc", "repository_url": "https://github.com/mlcommons/inference_results_v6.0", "details_url": "https://github.com/mlcommons/inference_results_v6.0/blob/4d3916ac9cf474b679cdfcf492d43a0559418ad1/summary_results.json", "submission_details_url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Nebius/results/nebius_b300_n1", "commit_sha": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "source_paths": [ "summary_results.json", "./closed/Nebius/results/nebius_b300_n1/qwen3-vl-235b-a22b/server/performance/run_1" ], "content_hashes": { "summary_results.json": "52fa813a27834e8de38eca9fd381688df1cf9cd95020d2bd9e951f6cc16384ae" } }, "compatibility": { "provider_id": "nebius", "provider_name": "Nebius", "accelerator_models": [ "B300", "B300-SXM" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Nebius normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "current_reference_economics": { "kind": "throughput", "gpuCount": 8, "clusterHourlyRate": 62.8, "throughputPerSecond": 45.1513, "unit": "Queries/s", "referenceUnitCount": 1000, "costPerReferenceUnit": 0.3863553085834615 } }, { "id": "deepseek-r1-offline", "category": "inference", "benchmark": "deepseek-r1", "scenario": "Offline", "dataset": null, "quality_target": "exact_match: 81.58614402917047 TOKENS_PER_SAMPLE: 3722.1116681859617", "system": { "submitter": "Nebius", "platform": "nebius_b300_n1", "name": "Nebius B300 n1 (8x B300-SXM-270GB, TensorRT)", "status": "available", "division": "closed", "suite": "datacenter", "accelerator": "NVIDIA B300-SXM-270GB", "accelerator_count": 8, "software": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0" }, "measurement": { "metric": "throughput", "value": 69318.9, "unit": "Tokens/s", "aggregation": "official_reported_result", "run_count": 1, "raw_values": [ 69318.9 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/inference-datacenter/", "rules_url": "https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc", "repository_url": "https://github.com/mlcommons/inference_results_v6.0", "details_url": "https://github.com/mlcommons/inference_results_v6.0/blob/4d3916ac9cf474b679cdfcf492d43a0559418ad1/summary_results.json", "submission_details_url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Nebius/results/nebius_b300_n1", "commit_sha": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "source_paths": [ "summary_results.json", "./closed/Nebius/results/nebius_b300_n1/deepseek-r1/Offline/performance/run_1" ], "content_hashes": { "summary_results.json": "52fa813a27834e8de38eca9fd381688df1cf9cd95020d2bd9e951f6cc16384ae" } }, "compatibility": { "provider_id": "nebius", "provider_name": "Nebius", "accelerator_models": [ "B300", "B300-SXM" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Nebius normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "current_reference_economics": { "kind": "throughput", "gpuCount": 8, "clusterHourlyRate": 62.8, "throughputPerSecond": 69318.9, "unit": "Tokens/s", "referenceUnitCount": 1000000, "costPerReferenceUnit": 0.2516549518882216 } }, { "id": "deepseek-r1-server", "category": "inference", "benchmark": "deepseek-r1", "scenario": "Server", "dataset": null, "quality_target": "exact_match: 81.58614402917047 TOKENS_PER_SAMPLE: 3721.7894257064722", "system": { "submitter": "Nebius", "platform": "nebius_b300_n1", "name": "Nebius B300 n1 (8x B300-SXM-270GB, TensorRT)", "status": "available", "division": "closed", "suite": "datacenter", "accelerator": "NVIDIA B300-SXM-270GB", "accelerator_count": 8, "software": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0" }, "measurement": { "metric": "throughput", "value": 60413.4, "unit": "Tokens/s", "aggregation": "official_reported_result", "run_count": 1, "raw_values": [ 60413.4 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/inference-datacenter/", "rules_url": "https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc", "repository_url": "https://github.com/mlcommons/inference_results_v6.0", "details_url": "https://github.com/mlcommons/inference_results_v6.0/blob/4d3916ac9cf474b679cdfcf492d43a0559418ad1/summary_results.json", "submission_details_url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Nebius/results/nebius_b300_n1", "commit_sha": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "source_paths": [ "summary_results.json", "./closed/Nebius/results/nebius_b300_n1/deepseek-r1/Server/performance/run_1" ], "content_hashes": { "summary_results.json": "52fa813a27834e8de38eca9fd381688df1cf9cd95020d2bd9e951f6cc16384ae" } }, "compatibility": { "provider_id": "nebius", "provider_name": "Nebius", "accelerator_models": [ "B300", "B300-SXM" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Nebius normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "current_reference_economics": { "kind": "throughput", "gpuCount": 8, "clusterHourlyRate": 62.8, "throughputPerSecond": 60413.4, "unit": "Tokens/s", "referenceUnitCount": 1000000, "costPerReferenceUnit": 0.2887512446649989 } }, { "id": "gpt-oss-120b-offline", "category": "inference", "benchmark": "gpt-oss-120b", "scenario": "Offline", "dataset": null, "quality_target": "exact_match: 82.959", "system": { "submitter": "Nebius", "platform": "nebius_b300_n1", "name": "Nebius B300 n1 (8x B300-SXM-270GB, TensorRT)", "status": "available", "division": "closed", "suite": "datacenter", "accelerator": "NVIDIA B300-SXM-270GB", "accelerator_count": 8, "software": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0" }, "measurement": { "metric": "throughput", "value": 106885, "unit": "Tokens/s", "aggregation": "official_reported_result", "run_count": 1, "raw_values": [ 106885 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/inference-datacenter/", "rules_url": "https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc", "repository_url": "https://github.com/mlcommons/inference_results_v6.0", "details_url": "https://github.com/mlcommons/inference_results_v6.0/blob/4d3916ac9cf474b679cdfcf492d43a0559418ad1/summary_results.json", "submission_details_url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Nebius/results/nebius_b300_n1", "commit_sha": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "source_paths": [ "summary_results.json", "./closed/Nebius/results/nebius_b300_n1/gpt-oss-120b/offline/performance/run_1" ], "content_hashes": { "summary_results.json": "52fa813a27834e8de38eca9fd381688df1cf9cd95020d2bd9e951f6cc16384ae" } }, "compatibility": { "provider_id": "nebius", "provider_name": "Nebius", "accelerator_models": [ "B300", "B300-SXM" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Nebius normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "current_reference_economics": { "kind": "throughput", "gpuCount": 8, "clusterHourlyRate": 62.8, "throughputPerSecond": 106885, "unit": "Tokens/s", "referenceUnitCount": 1000000, "costPerReferenceUnit": 0.16320760110814841 } }, { "id": "gpt-oss-120b-server", "category": "inference", "benchmark": "gpt-oss-120b", "scenario": "Server", "dataset": null, "quality_target": "exact_match: 83.337", "system": { "submitter": "Nebius", "platform": "nebius_b300_n1", "name": "Nebius B300 n1 (8x B300-SXM-270GB, TensorRT)", "status": "available", "division": "closed", "suite": "datacenter", "accelerator": "NVIDIA B300-SXM-270GB", "accelerator_count": 8, "software": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0" }, "measurement": { "metric": "throughput", "value": 100437, "unit": "Tokens/s", "aggregation": "official_reported_result", "run_count": 1, "raw_values": [ 100437 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/inference-datacenter/", "rules_url": "https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc", "repository_url": "https://github.com/mlcommons/inference_results_v6.0", "details_url": "https://github.com/mlcommons/inference_results_v6.0/blob/4d3916ac9cf474b679cdfcf492d43a0559418ad1/summary_results.json", "submission_details_url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Nebius/results/nebius_b300_n1", "commit_sha": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "source_paths": [ "summary_results.json", "./closed/Nebius/results/nebius_b300_n1/gpt-oss-120b/server/performance/run_1" ], "content_hashes": { "summary_results.json": "52fa813a27834e8de38eca9fd381688df1cf9cd95020d2bd9e951f6cc16384ae" } }, "compatibility": { "provider_id": "nebius", "provider_name": "Nebius", "accelerator_models": [ "B300", "B300-SXM" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Nebius normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "4d3916ac9cf474b679cdfcf492d43a0559418ad1", "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "current_reference_economics": { "kind": "throughput", "gpuCount": 8, "clusterHourlyRate": 62.8, "throughputPerSecond": 100437, "unit": "Tokens/s", "referenceUnitCount": 1000000, "costPerReferenceUnit": 0.1736854390756837 } }, { "id": "coreweave-b200-64-gpt-oss-20b-training", "category": "training", "benchmark": "GPT-OSS 20B pretraining", "scenario": "Time to train to quality target", "dataset": "C4", "quality_target": "3.34 log perplexity", "system": { "submitter": "CoreWeave", "platform": "CoreWeave_B200_8x8", "name": "CoreWeave_B200_8x8", "status": "Available cloud", "division": "closed", "suite": "training", "accelerator": "NVIDIA Blackwell GPU (B200-SXM-180GB)", "accelerator_count": 64, "software": "PyTorch NVIDIA Release 26.04" }, "measurement": { "metric": "time_to_quality_target", "value": 1618.959125, "unit": "seconds", "aggregation": "trimmed_mean_discard_lowest_and_highest", "run_count": 10, "raw_values": [ 1633.455, 1607.037, 1618.683, 1629.369, 1600.467, 1612.618, 1625.308, 1615.127, 1625.956, 1617.575 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/training/", "repository_url": "https://github.com/mlcommons/training_results_v6.0", "details_url": "https://github.com/mlcommons/training_results_v6.0/tree/eabf23a07b2a0c60a289ff871dc3a46fff0d0421/CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b", "commit_sha": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "source_paths": [ "CoreWeave/systems/CoreWeave_B200_8x8.json", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_0.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_1.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_2.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_3.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_4.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_5.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_6.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_7.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_8.txt", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_9.txt" ], "content_hashes": { "CoreWeave/systems/CoreWeave_B200_8x8.json": "ee0ce7e22bca8ee1b03924e4664c1077db016313054102b880a046605604709c", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_0.txt": "d20f3653d948e4dd470fe1c985db39738ce560c4d0ac664a24c766770190fad3", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_1.txt": "cb38e936fb9e0a56fcf210fc123e980ec4a0720114045f66262656e9ad2d3c0a", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_2.txt": "cfe1849701670c5a658692160d7fabe3d3f121eb54217c648ec48a82e8780d4c", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_3.txt": "61acf85abd20e92dfd64d2a58be77fca765f75b397a4efdacd417d07632793fa", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_4.txt": "c939d327fce37eee3067a6ba3c92a8dd64b9284d1502c37aa9a8751635889d63", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_5.txt": "55bba85bb1a13a817443691e31e99235a8e99bc534d676ba5dd6915e6407074c", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_6.txt": "5e5137125c259b386dbc65cc4faa0f396f2fe23c8af7b5393ec1ee510514269a", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_7.txt": "2a7f8c90e29a5d7e7e26c08e12cddc3286f554b3413616a47ebb7ce86ccc2b1c", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_8.txt": "d79b7b4dd3a350fcca1e908988239e8acc62d99b793aa0208552034ff66dd29d", "CoreWeave/results/CoreWeave_B200_8x8/gpt_oss_20b/result_9.txt": "255eeb43c2854faf4529147bc7c6c73e4159e9502d340e1898fba4f4811cffd4" } }, "compatibility": { "provider_id": "coreweave", "provider_name": "CoreWeave", "accelerator_models": [ "B200" ], "accelerator_count": 64, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 64-GPU system.", "The current CoreWeave normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "price_context": { "matched_offer_id": "coreweave:b200:global:on-demand", "provider_id": "coreweave", "provider_name": "CoreWeave", "accelerator_model": "B200", "accelerator_variant": null, "region_scope": "global_or_unspecified", "regions": [], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 8.6, "per_gpu_hour_min": 8.6, "per_gpu_hour_max": 8.6, "required_gpus": 64, "cluster_hourly_rate": 550.4, "cluster_hourly_rate_min": 550.4, "cluster_hourly_rate_max": 550.4, "observed_at": "2026-08-13T04:53:37Z", "freshness_status": "recent", "source_url": "https://www.coreweave.com/pricing", "evidence_hash": "1d95e76dce83326c305bbf12abf1f36bb38a5bc1c4306dccbf3ebcdc1ebf9111", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized CoreWeave price is per GPU-hour; the benchmark system requires 64 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Region is not available in the current source.", "Live capacity is not available in the current source.", "CPU, RAM, storage and network inclusions are not normalized yet.", "The official source is an 8-GPU node; the displayed rate is normalized per GPU." ] }, "current_reference_economics": { "kind": "job", "gpuCount": 64, "clusterHourlyRate": 550.4, "durationSeconds": 1618.959125, "durationHours": 0.4497108680555556, "jobCost": 247.52086177777778 } }, { "id": "coreweave-b200-64-llama31-8b-training", "category": "training", "benchmark": "Llama 3.1 8B pretraining", "scenario": "Time to train to quality target", "dataset": "C4", "quality_target": "3.3 log perplexity", "system": { "submitter": "CoreWeave", "platform": "CoreWeave_B200_8x8", "name": "CoreWeave_B200_8x8", "status": "Available cloud", "division": "closed", "suite": "training", "accelerator": "NVIDIA Blackwell GPU (B200-SXM-180GB)", "accelerator_count": 64, "software": "PyTorch NVIDIA Release 26.04" }, "measurement": { "metric": "time_to_quality_target", "value": 992.176875, "unit": "seconds", "aggregation": "trimmed_mean_discard_lowest_and_highest", "run_count": 10, "raw_values": [ 971.869, 971.916, 971.987, 972.02, 972.123, 979.816, 1022.96, 1023.202, 1023.391, 1074.075 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/training/", "repository_url": "https://github.com/mlcommons/training_results_v6.0", "details_url": "https://github.com/mlcommons/training_results_v6.0/tree/eabf23a07b2a0c60a289ff871dc3a46fff0d0421/CoreWeave/results/CoreWeave_B200_8x8/llama31_8b", "commit_sha": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "source_paths": [ "CoreWeave/systems/CoreWeave_B200_8x8.json", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_0.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_1.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_2.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_3.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_4.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_5.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_6.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_7.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_8.txt", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_9.txt" ], "content_hashes": { "CoreWeave/systems/CoreWeave_B200_8x8.json": "ee0ce7e22bca8ee1b03924e4664c1077db016313054102b880a046605604709c", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_0.txt": "f52d71ee8235a3b878490db4d0363e411eb85d34f802c3960f89c587adf6788d", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_1.txt": "80180f5615bee7c286f776ea4755ebe7096a80153f01b753dd367ed15d82d7e9", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_2.txt": "9b249de333b2f887ff31d52525d9e0a00230b4aa782f4c57bf8423b0148cbcc5", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_3.txt": "ce3d89fbe17b6f9c3f80ad86670e32b4ba5c0f8f21fb0fbfb5a6bbdc3605f4ae", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_4.txt": "1a12ec7a2bab480aade951f1ae61cafca6f9476e58c14a3e9089b88cd1f5d2d4", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_5.txt": "57c046e38c9fb9f5bdf466f6e507a589eb720bad60933e51716672d84aa8223b", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_6.txt": "3a332894fecd6d654c67f234fe7f91a48f6c5bcf8039a540277cb20164077e0d", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_7.txt": "01bd72b0954ba4a9071e4dd78d92936616e84767d9332e4cb3d964e5bc157d55", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_8.txt": "96a0a73e9db9a91dfdd4ebae8f6b43130b611cc90e1e8de347f9a0e36c62bb8e", "CoreWeave/results/CoreWeave_B200_8x8/llama31_8b/result_9.txt": "db68682aef7f30f9c0d3b27ec404bec3007c7fbad68bf00aefbeae6d021ba2ac" } }, "compatibility": { "provider_id": "coreweave", "provider_name": "CoreWeave", "accelerator_models": [ "B200" ], "accelerator_count": 64, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 64-GPU system.", "The current CoreWeave normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "price_context": { "matched_offer_id": "coreweave:b200:global:on-demand", "provider_id": "coreweave", "provider_name": "CoreWeave", "accelerator_model": "B200", "accelerator_variant": null, "region_scope": "global_or_unspecified", "regions": [], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 8.6, "per_gpu_hour_min": 8.6, "per_gpu_hour_max": 8.6, "required_gpus": 64, "cluster_hourly_rate": 550.4, "cluster_hourly_rate_min": 550.4, "cluster_hourly_rate_max": 550.4, "observed_at": "2026-08-13T04:53:37Z", "freshness_status": "recent", "source_url": "https://www.coreweave.com/pricing", "evidence_hash": "1d95e76dce83326c305bbf12abf1f36bb38a5bc1c4306dccbf3ebcdc1ebf9111", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized CoreWeave price is per GPU-hour; the benchmark system requires 64 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Region is not available in the current source.", "Live capacity is not available in the current source.", "CPU, RAM, storage and network inclusions are not normalized yet.", "The official source is an 8-GPU node; the displayed rate is normalized per GPU." ] }, "current_reference_economics": { "kind": "job", "gpuCount": 64, "clusterHourlyRate": 550.4, "durationSeconds": 992.176875, "durationHours": 0.2756046875, "jobCost": 151.69281999999998 } }, { "id": "gpt-oss-20b-training", "category": "training", "benchmark": "GPT-OSS 20B pretraining", "scenario": "Time to train to quality target", "dataset": "C4", "quality_target": "3.34 log perplexity", "system": { "submitter": "Nebius", "platform": "nebius_b300_n1", "name": "Nebius B300 n1 (8x B300-SXM-270GB)", "status": "Available cloud", "division": "closed", "suite": "training", "accelerator": "NVIDIA Blackwell Ultra GPU (B300-SXM-270GB)", "accelerator_count": 8, "software": "NVIDIA NeMo Framework Release 25.09" }, "measurement": { "metric": "time_to_quality_target", "value": 5008.8865, "unit": "seconds", "aggregation": "trimmed_mean_discard_lowest_and_highest", "run_count": 10, "raw_values": [ 4934.568, 5225.219, 5221.922, 4932.691, 5222.978, 4936.958, 4940.877, 4935.213, 4937.695, 4940.881 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/training/", "repository_url": "https://github.com/mlcommons/training_results_v6.0", "details_url": "https://github.com/mlcommons/training_results_v6.0/tree/eabf23a07b2a0c60a289ff871dc3a46fff0d0421/Nebius/results/nebius_b300_n1/gpt_oss_20b", "commit_sha": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "source_paths": [ "Nebius/systems/nebius_b300_n1.json", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_01.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_02.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_03.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_04.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_05.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_06.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_07.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_08.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_09.txt", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_10.txt" ], "content_hashes": { "Nebius/systems/nebius_b300_n1.json": "6db38cb0c2de0e3cea92b4e134dd6ead67e8c017725c027646abf5a2e1fbbc41", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_01.txt": "198e7cbe4c19907000258cb3303f5a8b213f28cf6b6370b401e88b34e2f74f64", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_02.txt": "d87f6d1f46a0d2afe788b4c509470dfd451d251554083bc5f42036a7da256c2c", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_03.txt": "ff14f1129c73f16fb417ec8cb2492953eea948b5118033c8ec784a9b133739d6", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_04.txt": "c13244d0b69458534e36e2eda12c23a658f73a76aa855a1db7a11269c87c79fb", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_05.txt": "9fc9c87fad97d7d4974d65de7bdab8d0d532e3c336c8d118b1b8a5de2e5a7cf9", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_06.txt": "0d7bb2022fdeacc4fd941ed2186702fd211fbfbcc6389d1d257056f0da6f5149", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_07.txt": "ff9ff37a95f980de9341bb3ac00953ae5c52fd085b6367c14631f1a1b3866f5d", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_08.txt": "537f1503033047aa218ca5f7471a63dd83f58bdfea64489403f6661e4a8c93d6", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_09.txt": "0e8903192d5216a1bcfa1c1fd2607f8a38ba12b7f7b809ee5b3d54b66610971a", "Nebius/results/nebius_b300_n1/gpt_oss_20b/result_10.txt": "722c50b8d919bfa26afb2b9d2943a8d4de68dccd90afa0ebe50724e0005c28b6" } }, "compatibility": { "provider_id": "nebius", "provider_name": "Nebius", "accelerator_models": [ "B300", "B300-SXM" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Nebius normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "current_reference_economics": { "kind": "job", "gpuCount": 8, "clusterHourlyRate": 62.8, "durationSeconds": 5008.8865, "durationHours": 1.391357361111111, "jobCost": 87.37724227777777 } }, { "id": "lambda-b200-8-gpt-oss-20b-training", "category": "training", "benchmark": "GPT-OSS 20B pretraining", "scenario": "Time to train to quality target", "dataset": "C4", "quality_target": "3.34 log perplexity", "system": { "submitter": "Lambda", "platform": "b200_n1_ngc25.09_nemo", "name": "Lambda-1-Click-Cluster_B200_n1", "status": "Available cloud", "division": "closed", "suite": "training", "accelerator": "NVIDIA Blackwell GPU (B200-SXM-180GB)", "accelerator_count": 8, "software": "PyTorch NVIDIA Release 25.04" }, "measurement": { "metric": "time_to_quality_target", "value": 5787.4255, "unit": "seconds", "aggregation": "trimmed_mean_discard_lowest_and_highest", "run_count": 10, "raw_values": [ 5744.792, 5741.946, 6101.439, 5739.036, 5390.648, 6101.844, 5738.169, 5709.949, 5765.243, 5758.83 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/training/", "repository_url": "https://github.com/mlcommons/training_results_v6.0", "details_url": "https://github.com/mlcommons/training_results_v6.0/tree/eabf23a07b2a0c60a289ff871dc3a46fff0d0421/Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b", "commit_sha": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "source_paths": [ "Lambda/systems/b200_n1_ngc25.09_nemo.json", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_1.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_2.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_3.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_4.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_5.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_6.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_7.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_8.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_9.txt", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_10.txt" ], "content_hashes": { "Lambda/systems/b200_n1_ngc25.09_nemo.json": "1a8b436324651065fab3c9673b8c321a6151a1dfb00dbd18399ee59db5152366", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_1.txt": "560df338a4ea7daf99a54b7ccf9c9ab5ea521e648a3efc4099617353361815cb", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_2.txt": "77a12dae3e9a8017ae3f0fc46efa788101a2fff49499a80afb816bd999f94521", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_3.txt": "27f8614182a7149605882212acf2e583c7e4babdbc12a4018cd51eabf74f8e3d", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_4.txt": "9d801b569924bffc0fc48800a720eca1dec564b6c753e53560c7599393f51e49", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_5.txt": "2c0f5262f21210ba103987b67c20e9a9673e7c345e70e0870730841cbb1ae82f", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_6.txt": "0129aaff1eae5cdd72219dd6e21fd20a2fd9a4f9f2ee5f56f2149b16baaa4460", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_7.txt": "4aaccecdce029e7c0d3d34e2f397c3d1154d0b0af488186dc057aa8907fb398d", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_8.txt": "a92b73b70a9d7e18fa75d7ef7a607df00fe79504900ccc17735153773a1d9629", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_9.txt": "316ee27d9550b2427855fa0bf0db926e854554444cce08ccafcca21ec6aee5ef", "Lambda/results/b200_n1_ngc25.09_nemo/gpt_oss_20b/result_10.txt": "b59d44b070d6b04c2234293d5d0923ab7ed00ecaf36ee6cfdd7edbd87ed0eb75" } }, "compatibility": { "provider_id": "lambda", "provider_name": "Lambda", "accelerator_models": [ "B200" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Lambda normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "price_context": { "matched_offer_id": "lambda:b200:global:on-demand", "provider_id": "lambda", "provider_name": "Lambda", "accelerator_model": "B200", "accelerator_variant": null, "region_scope": "global_or_unspecified", "regions": [], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 6.69, "per_gpu_hour_min": 6.69, "per_gpu_hour_max": 6.69, "required_gpus": 8, "cluster_hourly_rate": 53.52, "cluster_hourly_rate_min": 53.52, "cluster_hourly_rate_max": 53.52, "observed_at": "2026-08-13T04:53:37Z", "freshness_status": "recent", "source_url": "https://lambda.ai/pricing", "evidence_hash": "1621fbbdc19655ea81fda2416b5196b530d491dd82b5dc329a31b055b003f2ce", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Lambda price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Region is not available in the current source.", "Live capacity is not available in the current source.", "CPU, RAM, storage and network inclusions are not normalized yet." ] }, "current_reference_economics": { "kind": "job", "gpuCount": 8, "clusterHourlyRate": 53.52, "durationSeconds": 5787.4255, "durationHours": 1.6076181944444445, "jobCost": 86.03972576666668 } }, { "id": "lambda-b200-8-llama31-8b-training", "category": "training", "benchmark": "Llama 3.1 8B pretraining", "scenario": "Time to train to quality target", "dataset": "C4", "quality_target": "3.3 log perplexity", "system": { "submitter": "Lambda", "platform": "b200_n1_ngc25.09_nemo", "name": "Lambda-1-Click-Cluster_B200_n1", "status": "Available cloud", "division": "closed", "suite": "training", "accelerator": "NVIDIA Blackwell GPU (B200-SXM-180GB)", "accelerator_count": 8, "software": "PyTorch NVIDIA Release 25.04" }, "measurement": { "metric": "time_to_quality_target", "value": 5115.789375, "unit": "seconds", "aggregation": "trimmed_mean_discard_lowest_and_highest", "run_count": 10, "raw_values": [ 5025.99, 5384.276, 5025.616, 5024.626, 5384.869, 5024.409, 5027.064, 5026.72, 5027.154, 5386.938 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/training/", "repository_url": "https://github.com/mlcommons/training_results_v6.0", "details_url": "https://github.com/mlcommons/training_results_v6.0/tree/eabf23a07b2a0c60a289ff871dc3a46fff0d0421/Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b", "commit_sha": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "source_paths": [ "Lambda/systems/b200_n1_ngc25.09_nemo.json", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_1.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_2.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_3.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_4.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_5.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_6.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_7.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_8.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_9.txt", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_10.txt" ], "content_hashes": { "Lambda/systems/b200_n1_ngc25.09_nemo.json": "1a8b436324651065fab3c9673b8c321a6151a1dfb00dbd18399ee59db5152366", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_1.txt": "a7f671077b10de7122fca451562df639d1caa18210fda90fbbfbd97ad78a1b05", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_2.txt": "740525d607ef4ac6d4c54f4e57c7f68e7232555ad4cbdfc9fd0ba21259cf4fa0", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_3.txt": "69911b6d00ee67c4aa9d5cb512039a54cd7737bb2fe2e2612ac181e68b8606b5", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_4.txt": "9ee7f3578f00b27921eed0031b24feacd604b58e5bc4c3a17e0370418311f864", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_5.txt": "1a24ce6a36df37a9e354493f85ac1a3899e062661b8668165a4fe49990eac7c5", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_6.txt": "edbad6fca557f67cc894498f71681ce5d5745f3ac6095d09e32efa928c2fbbb4", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_7.txt": "05a25156c1e72511ed7f8300dde9cf7a820c8e5a1bbcd43b7b39df143d0cc61d", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_8.txt": "cdb590d0777919fdc852084085082f7849d15d05b32c09a735497dfc6e82d176", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_9.txt": "b2de65f173b90c8ac09bafa44815d1bec9735c64c2c871c6288221da9e1a7ef6", "Lambda/results/b200_n1_ngc25.09_nemo/llama31_8b/result_10.txt": "3b97b1c0dbde85cbe9ea44b5a999340972761f81dbcb5764a71101a0f9392538" } }, "compatibility": { "provider_id": "lambda", "provider_name": "Lambda", "accelerator_models": [ "B200" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Lambda normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "price_context": { "matched_offer_id": "lambda:b200:global:on-demand", "provider_id": "lambda", "provider_name": "Lambda", "accelerator_model": "B200", "accelerator_variant": null, "region_scope": "global_or_unspecified", "regions": [], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 6.69, "per_gpu_hour_min": 6.69, "per_gpu_hour_max": 6.69, "required_gpus": 8, "cluster_hourly_rate": 53.52, "cluster_hourly_rate_min": 53.52, "cluster_hourly_rate_max": 53.52, "observed_at": "2026-08-13T04:53:37Z", "freshness_status": "recent", "source_url": "https://lambda.ai/pricing", "evidence_hash": "1621fbbdc19655ea81fda2416b5196b530d491dd82b5dc329a31b055b003f2ce", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Lambda price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Region is not available in the current source.", "Live capacity is not available in the current source.", "CPU, RAM, storage and network inclusions are not normalized yet." ] }, "current_reference_economics": { "kind": "job", "gpuCount": 8, "clusterHourlyRate": 53.52, "durationSeconds": 5115.789375, "durationHours": 1.4210526041666667, "jobCost": 76.054735375 } }, { "id": "llama31-8b-training", "category": "training", "benchmark": "Llama 3.1 8B pretraining", "scenario": "Time to train to quality target", "dataset": "C4", "quality_target": "3.3 log perplexity", "system": { "submitter": "Nebius", "platform": "nebius_b300_n1", "name": "Nebius B300 n1 (8x B300-SXM-270GB)", "status": "Available cloud", "division": "closed", "suite": "training", "accelerator": "NVIDIA Blackwell Ultra GPU (B300-SXM-270GB)", "accelerator_count": 8, "software": "NVIDIA NeMo Framework Release 25.09" }, "measurement": { "metric": "time_to_quality_target", "value": 4320.74175, "unit": "seconds", "aggregation": "trimmed_mean_discard_lowest_and_highest", "run_count": 10, "raw_values": [ 4322.608, 4322.73, 4318.691, 4321.582, 4319.838, 4319.319, 4321.173, 4323.013, 4318.259, 4319.993 ] }, "evidence": { "source_type": "official_peer_reviewed_benchmark", "methodology_url": "https://mlcommons.org/benchmarks/training/", "repository_url": "https://github.com/mlcommons/training_results_v6.0", "details_url": "https://github.com/mlcommons/training_results_v6.0/tree/eabf23a07b2a0c60a289ff871dc3a46fff0d0421/Nebius/results/nebius_b300_n1/llama31_8b", "commit_sha": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "source_paths": [ "Nebius/systems/nebius_b300_n1.json", "Nebius/results/nebius_b300_n1/llama31_8b/result_01.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_02.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_03.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_04.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_05.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_06.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_07.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_08.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_09.txt", "Nebius/results/nebius_b300_n1/llama31_8b/result_10.txt" ], "content_hashes": { "Nebius/systems/nebius_b300_n1.json": "6db38cb0c2de0e3cea92b4e134dd6ead67e8c017725c027646abf5a2e1fbbc41", "Nebius/results/nebius_b300_n1/llama31_8b/result_01.txt": "f0ded7d9bf101c7dfd0514dad13ea32d538e482e5b8833150607ae7363dcf911", "Nebius/results/nebius_b300_n1/llama31_8b/result_02.txt": "381b2dba4eceb48221b519f1cef852f31e6e4d5b2ae75c4424baad0d1d514054", "Nebius/results/nebius_b300_n1/llama31_8b/result_03.txt": "07e41de985a3777ecad34723dcc2d35f8ff0c0872cb8a7c826e58ee189da15cc", "Nebius/results/nebius_b300_n1/llama31_8b/result_04.txt": "6caaa4f0e60ff6067dc63a490f80532503acdb46549c0cf3923e3117cf8f3058", "Nebius/results/nebius_b300_n1/llama31_8b/result_05.txt": "5073871dc3af83793728fee58e15334068b4e3f314d188a922b044515c1f025f", "Nebius/results/nebius_b300_n1/llama31_8b/result_06.txt": "0f00a3c5e7ab77c31dac6d30a64877604031c7d73a9152d1cd8d522edc1c8d53", "Nebius/results/nebius_b300_n1/llama31_8b/result_07.txt": "047180040326d005a21cc12f296a97ff1a24b444447fabefe4cb105de26aad34", "Nebius/results/nebius_b300_n1/llama31_8b/result_08.txt": "957cf0c0e13d139556d7b41c01379bd2eb61ac0a9b98ec269715628c3a81e365", "Nebius/results/nebius_b300_n1/llama31_8b/result_09.txt": "ccbc02e92a1272777fbb95e89616bbdcb414d7439cc0eac6de54549f2249a70b", "Nebius/results/nebius_b300_n1/llama31_8b/result_10.txt": "3dfe7331b6dc46464b0eb67c81ed857633f9e23436b7b1cb197135ac853e6441" } }, "compatibility": { "provider_id": "nebius", "provider_name": "Nebius", "accelerator_models": [ "B300", "B300-SXM" ], "accelerator_count": 8, "market_type": "on_demand", "pricing_unit": "gpu_hour", "match_scope": "same_provider_accelerator_model_and_market", "cost_projection": "normalized_per_gpu_rate_times_benchmark_accelerator_count" }, "limitations": [ "The result applies only to the published model, software, scenario and 8-GPU system.", "The current Nebius normalized per-GPU rate is joined separately and multiplied by the benchmark GPU count; identical topology and simultaneous capacity are not asserted.", "Storage, data transfer, orchestration, queueing, support, taxes and discounts are excluded.", "A benchmark result is not a guarantee of performance for a different workload or configuration." ], "source_commit": "eabf23a07b2a0c60a289ff871dc3a46fff0d0421", "price_context": { "matched_offer_id": "nebius:gpu-b300-sxm:eu-west2:on_demand", "provider_id": "nebius", "provider_name": "Nebius AI Cloud", "accelerator_model": "B300", "accelerator_variant": "NVLink 288GB", "region_scope": "regional", "regions": [ "eu-west2" ], "market_type": "on_demand", "currency": "USD", "per_gpu_hour": 7.85, "per_gpu_hour_min": 7.85, "per_gpu_hour_max": 7.85, "required_gpus": 8, "cluster_hourly_rate": 62.8, "cluster_hourly_rate_min": 62.8, "cluster_hourly_rate_max": 62.8, "observed_at": "2026-08-13T02:40:10Z", "freshness_status": "recent", "source_url": "https://docs.nebius.com/compute/resources/pricing", "evidence_hash": "67f8f1f1c2a7ce5ecc3fd3e5f413dcb7260f460711da4091b3bed4f0a8c9e9e8", "pricing_match_method": "same provider + exact normalized GPU model + on-demand market; current per-GPU rate multiplied by the published benchmark GPU count", "limitations": [ "The normalized Nebius AI Cloud price is per GPU-hour; the benchmark system requires 8 GPUs simultaneously.", "The price match does not prove that the benchmark topology, interconnect, software or simultaneous capacity is currently purchasable.", "Storage, network, orchestration, support, reservation, taxes and discounts are excluded unless the source row explicitly states otherwise.", "Storage, data transfer, support, taxes and optional services are excluded unless explicitly present in resources.", "The official pricing documentation proves regional platform availability, not immediate numeric stock." ] }, "current_reference_economics": { "kind": "job", "gpuCount": 8, "clusterHourlyRate": 62.8, "durationSeconds": 4320.74175, "durationHours": 1.2002060416666667, "jobCost": 75.37293941666667 } } ] }