MiniMax M2.5/M2.7 — B300 vs GB200 NVL72
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and GB200 NVL72 (NVIDIA Blackwell) on MiniMax M2.5/M2.7. 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.
Near the low end of the 31–180 tok/s/user interactivity band, at 68 tok/s/user on MiniMax M2.5/M2.7: B300 runs 7214 tok/s/chip at $0.09/M tokens, GB200 NVL72 runs 8433 at $0.06/M. GB200 NVL72 is 44% cheaper per token; GB200 NVL72 delivers 17% more tok/s/chip.
Setting 105 tok/s/user as the target on MiniMax M2.5/M2.7, B300 produces 3089 tok/s/chip ($0.20 per million tokens) and GB200 NVL72 produces 3247 ($0.16). GB200 NVL72 is 27% cheaper per token; GB200 NVL72 delivers 5% more tok/s/chip.
At 143 tok/s/user interactivity on MiniMax M2.5/M2.7, B300 delivers 1147 tok/s/chip at $0.55 per million tokens; GB200 NVL72 delivers 1066 tok/s/chip at $0.49. GB200 NVL72 is 12% cheaper per token; B300 delivers 8% more tok/s/chip at this point. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:7214.1GB200 NVL72:8432.8 | B300:3089.2GB200 NVL72:3246.9 | B300:1147.4GB200 NVL72:1066.4 |
| Cost ($/M tok) | B300:$0.088GB200 NVL72:$0.061 | B300:$0.203GB200 NVL72:$0.160 | B300:$0.547GB200 NVL72:$0.489 |
| tok/s/MW | B300:3796911GB200 NVL72:4509515 | B300:1625869GB200 NVL72:1736305 | B300:603898GB200 NVL72:570241 |
| Concurrency | B300:~558GB200 NVL72:~928 | B300:~146GB200 NVL72:~236 | B300:~7GB200 NVL72:~45 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.