LLM Inference Benchmark Explorer
MiMo-V2.6-Pro-RL on 8× DGX Spark — FP4, SGLang, TP8, speculative decoding k=7 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | MiMo-V2.6-Pro-RL |
|---|---|
| Parameters | 1.02T |
| Intelligence Index | 46.3 |
| Agentic Index | — |
| Device | 8× DGX Spark |
| Quantization | FP4 |
| Max C | 2 |
| Chat Capacity | 8 |
| Agentic Capacity | 3 |
| TP | 8 |
| DP | — |
| PP | — |
| Engine | SGLang |
| Speculative Decoding | Yes |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 372 | 38.83 | PASS |
| 2 | 461 | 26.35 | PASS |
| 4 | 660 | 17.52 | FAIL |
| 8 | 5242 | 7.61 | FAIL |
In OpenZeka's measurement, MiMo-V2.6-Pro-RL (1.02T parameters), served in FP4 format with SGLang and speculative decoding on 8× DGX Spark (TP=8), reached a generation speed of 38.8 tok/s per request and a time to first token (TTFT) of 372 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 8 users for chat use and 3 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, 256K context (max_model_len 262144), SGLang. Model: XiaomiMiMo/MiMo-V2.6-Pro-RL (served locally as mimo-v2.6-pro)
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, MiMo-V2.6-Pro-RL (1.02T parameters), served in FP4 format with SGLang and speculative decoding on 8× DGX Spark (TP=8), reached a generation speed of 38.8 tok/s per request and a time to first token (TTFT) of 372 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 8 users for chat use and 3 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.