GB300 NVL72CoreWeave · 64 accelerators
910,468tokens/s
↗MLPerf Inference v6.0Closed divisionOfficial results
Find the fastest verified inference hardware for your workload.
Not fastest in general. Fastest inside one exact, comparable benchmark slice—with every row linked to the source commit.
Current comparison
gpt-oss-120b · Offline
All rows use MLPerf Inference v6.0 Closed division, gpt-oss-120b, Offline, official-default accuracy evidence, and official submitted-system tokens/s. Per-accelerator values are derived and never the default.
Slice ID
v6.0-closed-gpt-oss-120b-offline-official-default-tokens-per-secondTop systems by accelerator family
View all 27 results →One metric. One unit. No blended claims.
Comparable or excluded
Release, division, workload, scenario, accuracy target, metric and unit must match. Ambiguity enters quarantine—not the ranking.
Systems before chips
The official submitted-system score is primary. Per-accelerator values appear only as labeled arithmetic when the source explicitly states count.
Source at every row
Repository, commit, path, URL and SHA-256 travel with each published record. The index can be rebuilt and audited without trusting this interface.
For autonomous research agents
Ask an exact question. Pay only for the exact answer.
Status and five-row previews are free. Full rankings cost $0.02; exact 2–8 chip comparisons cost $0.03 on Base Sepolia when payment is configured.
POST /api/agent/entrypoints/
rank-inference-chips/invokeOpen API guide