Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB200 NVL72 (NVIDIA Blackwell) 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 →
Near the low end of the 65–398 tok/s/user interactivity band, at 148 tok/s/user on MiniMax M3 428B: B200 runs 37671 tok/s/chip at $0.01/M tokens, GB200 NVL72 runs 29017 at $0.02/M. B200 is 40% cheaper per token; B200 delivers 30% more tok/s/chip.
Setting 232 tok/s/user as the target on MiniMax M3 428B, B200 produces 19470 tok/s/chip ($0.02 per million tokens) and GB200 NVL72 produces 16290 ($0.03). B200 is 28% cheaper per token; B200 delivers 20% more tok/s/chip.
At 315 tok/s/user interactivity on MiniMax M3 428B, B200 delivers 6841 tok/s/chip at $0.07 per million tokens; GB200 NVL72 delivers 7328 tok/s/chip at $0.07. Cost per token is essentially tied; GB200 NVL72 delivers 7% more tok/s/chip at this point. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:37670.6GB200 NVL72:29016.9 | B200:19469.7GB200 NVL72:16290.4 | B200:6840.7GB200 NVL72:7328.2 |
| Cost ($/M tok) | B200:$0.013GB200 NVL72:$0.018 | B200:$0.025GB200 NVL72:$0.032 | B200:$0.070GB200 NVL72:$0.071 |
| tok/s/MW | B200:22029581GB200 NVL72:15517059 | B200:11385783GB200 NVL72:8711435 | B200:4000418GB200 NVL72:3918832 |
| Concurrency | B200:~22GB200 NVL72:~17 | B200:~12GB200 NVL72:~9 | B200:~2GB200 NVL72:~2 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
TCO $/chip/hr:
Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate cost per million tokens per decode chip or per prefill chip, rather than per total chip count. This makes direct token cost comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate input throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct input throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate output throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct output throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate power per decode chip or per prefill chip, rather than per total chip count. This makes direct power comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate Joules per decode chip or per prefill chip, rather than per total chip count. This makes direct Joules per token comparison with aggregated configs not an apples-to-apples comparison.
Matching measurements exist, but their chip series are hidden.
Shift+Scroll to zoom • Drag to pan • Double-click to reset • Click a point to pin tooltip