MiniMax M3 428B — GB300 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on MiniMax M3 428B. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
AgentX replays real coding-agent sessions rather than fixed-length prompts, so context grows turn over turn and most of each request is served from cache instead of being recomputed. That turns the comparison into a systems question: KV transfer between nodes, prefix-aware routing, and cache capacity all move the curve alongside raw chip throughput. Fixed-sequence workloads stay the clean baseline for kernel and silicon performance, so the two scenarios answer different questions about the same hardware. Learn more about AgentX →
H200 hits 7815 tok/s/chip for $0.04 per million tokens at 141 tok/s/user on MiniMax M3 428B. No GB300 NVL72 data at this operating point.
H200: 4685 tok/s/chip, $0.07 per million tokens at 188 tok/s/user on MiniMax M3 428B. GB300 NVL72 is unmeasured here.
At 235 tok/s/user on MiniMax M3 428B, H200 delivers 2926 tok/s/chip at $0.12 per million tokens; GB300 NVL72 hasn't been benchmarked at this target. (Numbers reflect the default agentic-traces · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:—H200:7815.3 | GB300 NVL72:—H200:4684.7 | GB300 NVL72:—H200:2926.1 |
| Cost ($/M tok) | GB300 NVL72:—H200:$0.043 | GB300 NVL72:—H200:$0.072 | GB300 NVL72:—H200:$0.116 |
| tok/s/MW | GB300 NVL72:—H200:5704564 | GB300 NVL72:—H200:3419513 | GB300 NVL72:—H200:2135827 |
| Concurrency | GB300 NVL72:—H200:~11 | GB300 NVL72:—H200:~5 | GB300 NVL72:—H200:~3 |