MiniMax M3 428B — GB300 NVL72 vs H100
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H100 (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 →
H100: 780 tok/s/chip, $0.42 per million tokens at 7 tok/s/user on MiniMax M3 428B. GB300 NVL72 is unmeasured here.
At 11 tok/s/user on MiniMax M3 428B, H100 delivers 659 tok/s/chip at $0.49 per million tokens; GB300 NVL72 hasn't been benchmarked at this target.
H100 hits 637 tok/s/chip for $0.51 per million tokens at 15 tok/s/user on MiniMax M3 428B. No GB300 NVL72 data at this operating point. (Numbers reflect the default agentic-traces · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:—H100:780.2 | GB300 NVL72:—H100:659.0 | GB300 NVL72:—H100:637.0 |
| Cost ($/M tok) | GB300 NVL72:—H100:$0.417 | GB300 NVL72:—H100:$0.493 | GB300 NVL72:—H100:$0.510 |
| tok/s/MW | GB300 NVL72:—H100:569485 | GB300 NVL72:—H100:481042 | GB300 NVL72:—H100:464987 |
| Concurrency | GB300 NVL72:—H100:~6 | GB300 NVL72:—H100:~3 | GB300 NVL72:—H100:~1 |