LLM Inference Benchmark Explorer
Kimi-K3.1-Pro on DGX B300 — MXFP4, SGLang, TP8, speculative decoding k=8 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | Kimi-K3.1-Pro |
|---|---|
| Parameters | 2.78T |
| Intelligence Index | — |
| Agentic Index | — |
| Device | DGX B300 |
| Quantization | MXFP4 |
| Max C | 16 |
| Chat Capacity | 64 |
| Agentic Capacity | 24 |
| TP | 8 |
| DP | — |
| PP | — |
| Engine | SGLang |
| Speculative Decoding | Yes |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 379 | 176.03 | PASS |
| 2 | 403 | 114.49 | PASS |
| 4 | 553 | 70.71 | PASS |
| 8 | 596 | 44.91 | PASS |
| 16 | 727 | 27.57 | PASS |
| 32 | 1365 | 35.40 | FAIL |
| 64 | 5081 | 18.11 | FAIL |
| 128 | 12519 | 9.55 | FAIL |
In OpenZeka's measurement, Kimi-K3.1-Pro (2.78T parameters), served in MXFP4 format with SGLang and speculative decoding on DGX B300 (TP=8), reached a generation speed of 176 tok/s per request and a time to first token (TTFT) of 379 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 64 users for chat use and 24 for agentic use. The realistic capacity will likely fall between these two values. In scenarios dominated by coding, tool use, long workflows and multi-agent use, capacity approaches the agentic estimate; where shorter interactions, standard conversations and lighter tasks dominate, it approaches the chat estimate.
Notes: DFlash2 k=8, 1M context. Multimodal checkpoint measured on text. Model: moonshotai/Kimi-K3.1-Pro
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, Kimi-K3.1-Pro (2.78T parameters), served in MXFP4 format with SGLang and speculative decoding on DGX B300 (TP=8), reached a generation speed of 176 tok/s per request and a time to first token (TTFT) of 379 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 64 users for chat use and 24 for agentic use. The realistic capacity will likely fall between these two values. In scenarios dominated by coding, tool use, long workflows and multi-agent use, capacity approaches the agentic estimate; where shorter interactions, standard conversations and lighter tasks dominate, it approaches the chat estimate.