Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) on DeepSeek V4.1 Flash 552B. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics β 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 β
Setting 146 tok/s/user as the target on DeepSeek V4.1 Flash 552B, GB200 NVL72 produces 104151 tok/s/chip ($0.00 per million tokens) and GB300 NVL72 produces 732212 ($0.00). GB300 NVL72 is 466% cheaper per token; GB300 NVL72 delivers 603% more tok/s/chip.
At 269 tok/s/user interactivity on DeepSeek V4.1 Flash 552B, GB200 NVL72 delivers 20684 tok/s/chip at $0.02 per million tokens; GB300 NVL72 delivers 187624 tok/s/chip at $0.00. GB300 NVL72 is 630% cheaper per token; GB300 NVL72 delivers 807% more tok/s/chip at this point.
GB200 NVL72 posts 16190 tok/s/chip for $0.03 per million tokens at 392 tok/s/user on DeepSeek V4.1 Flash 552B; GB300 NVL72 posts 17879 tok/s/chip for $0.04. GB200 NVL72 is 12% cheaper per token; GB300 NVL72 delivers 10% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:104150.8GB300 NVL72:732211.8 | GB200 NVL72:20684.1GB300 NVL72:187623.7 | GB200 NVL72:16190.2GB300 NVL72:17879.3 |
| Cost ($/M tok) | GB200 NVL72:$0.005GB300 NVL72:$0.001 | GB200 NVL72:$0.025GB300 NVL72:$0.003 | GB200 NVL72:$0.032GB300 NVL72:$0.036 |
| tok/s/MW | GB200 NVL72:55695602GB300 NVL72:345382939 | GB200 NVL72:11061000GB300 NVL72:88501729 | GB200 NVL72:8657835GB300 NVL72:8433647 |
| Concurrency | GB200 NVL72:~55GB300 NVL72:~4096 | GB200 NVL72:~6GB300 NVL72:~618 | GB200 NVL72:~8GB300 NVL72:~10 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
No measurements to plot for this selection. Review the benchmark controls above or adjust quick filters.
