Qwen 3.8 Flash Next 176B-A6B — H100 vs H200
Head-to-head AI inference benchmark comparison of H100 (NVIDIA Hopper) and H200 (NVIDIA Hopper) 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 44 tok/s/user as the target on Qwen 3.8 Flash Next 176B-A6B, H100 produces 4983 tok/s/chip ($0.07 per million tokens) and H200 produces 16508 ($0.02). H200 is 218% cheaper per token; H200 delivers 231% more tok/s/chip.
At 74 tok/s/user interactivity on Qwen 3.8 Flash Next 176B-A6B, H100 delivers 4947 tok/s/chip at $0.07 per million tokens; H200 delivers 15464 tok/s/chip at $0.02. H200 is 200% cheaper per token; H200 delivers 213% more tok/s/chip at this point.
H100 posts 4914 tok/s/chip for $0.07 per million tokens at 104 tok/s/user on Qwen 3.8 Flash Next 176B-A6B; H200 posts 13308 tok/s/chip for $0.03. H200 is 160% cheaper per token; H200 delivers 171% more tok/s/chip. (Numbers reflect the default agentic-traces · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | H100:4983.0H200:16508.0 | H100:4947.0H200:15464.4 | H100:4913.9H200:13308.3 |
| Cost ($/M tok) | H100:$0.065H200:$0.021 | H100:$0.066H200:$0.022 | H100:$0.066H200:$0.025 |
| tok/s/MW | H100:3637239H200:12049649 | H100:3610925H200:11287849 | H100:3586797H200:9714104 |
| Concurrency | H100:~11H200:~15 | H100:~10H200:~13 | H100:~9H200:~11 |
Inference Performance
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.