LLM Inference Benchmark Explorer
MiMo-V2.6-Flash-RL on 4× DGX Spark — FP4, vLLM, TP4, speculative decoding k=7 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | MiMo-V2.6-Flash-RL |
|---|---|
| Parameters | 310B |
| Intelligence Index | — |
| Agentic Index | — |
| Device | 4× DGX Spark |
| Quantization | FP4 |
| Max C | 4 |
| Chat Capacity | 16 |
| Agentic Capacity | 6 |
| TP | 4 |
| DP | — |
| PP | — |
| Engine | vLLM |
| Speculative Decoding | Yes |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 303 | 45.68 | PASS |
| 2 | 402 | 32.47 | PASS |
| 4 | 497 | 22.94 | PASS |
| 8 | 643 | 14.88 | FAIL |
In OpenZeka's measurement, MiMo-V2.6-Flash-RL (310B parameters), served in FP4 format with vLLM and speculative decoding on 4× DGX Spark (TP=4), reached a generation speed of 45.7 tok/s per request and a time to first token (TTFT) of 303 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 16 users for chat use and 6 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: DFlash speculative decoding k=7, 500K context. Served as MiMo-V2.6-Flash-RL. Model: XiaomiMiMo/MiMo-V2.6-Flash-RL
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, MiMo-V2.6-Flash-RL (310B parameters), served in FP4 format with vLLM and speculative decoding on 4× DGX Spark (TP=4), reached a generation speed of 45.7 tok/s per request and a time to first token (TTFT) of 303 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 16 users for chat use and 6 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.