Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and B300 (NVIDIA Blackwell) on Qwen 3.8 Flash Next 176B-A6B. 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 →
Setting 82 tok/s/user as the target on Qwen 3.8 Flash Next 176B-A6B, B200 produces 63055 tok/s/chip ($0.01 per million tokens) and B300 produces 81505 ($0.01). B200 is 1% cheaper per token; B300 delivers 29% more tok/s/chip.
At 111 tok/s/user interactivity on Qwen 3.8 Flash Next 176B-A6B, B200 delivers 53999 tok/s/chip at $0.01 per million tokens; B300 delivers 65592 tok/s/chip at $0.01. B200 is 8% cheaper per token; B300 delivers 21% more tok/s/chip at this point.
B200 posts 42523 tok/s/chip for $0.01 per million tokens at 141 tok/s/user on Qwen 3.8 Flash Next 176B-A6B; B300 posts 50971 tok/s/chip for $0.01. B200 is 9% cheaper per token; B300 delivers 20% more tok/s/chip. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:63054.5B300:81505.3 | B200:53999.0B300:65591.9 | B200:42523.0B300:50970.9 |
| Cost ($/M tok) | B200:$0.008B300:$0.008 | B200:$0.009B300:$0.010 | B200:$0.011B300:$0.012 |
| tok/s/MW | B200:36874001B300:42897518 | B200:31578338B300:34522039 | B200:24867232B300:26826810 |
| Concurrency | B200:~14B300:~16 | B200:~12B300:~13 | B200:~9B300:~10 |
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.
No measurements to plot for this selection. Review the benchmark controls above or adjust quick filters.
Qwen3.8 Flash Next 176B