LLM Inference Benchmark Explorer
Mistral-Large-4-1T-A52B on DGX B300 — FP8, SGLang, TP8, speculative decoding k=3 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | Mistral-Large-4-1T-A52B |
|---|---|
| Parameters | 1T |
| Intelligence Index | 26.1 |
| Agentic Index | — |
| Device | DGX B300 |
| Quantization | FP8 |
| Max C | 128+ |
| Chat Capacity | 147 |
| Agentic Capacity | 36 |
| TP | 8 |
| DP | — |
| PP | — |
| Engine | SGLang |
| Speculative Decoding | Yes |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 108 | 157.36 | PASS |
| 2 | 148 | 124.75 | PASS |
| 4 | 181 | 102.46 | PASS |
| 8 | 205 | 77.36 | PASS |
| 16 | 220 | 64.96 | PASS |
| 32 | 286 | 50.14 | PASS |
| 64 | 306 | 37.76 | PASS |
| 128 | 347 | 30.36 | PASS |
In OpenZeka's measurement, Mistral-Large-4-1T-A52B (1T parameters), served in FP8 format with SGLang and speculative decoding on DGX B300 (TP=8), reached a generation speed of 157.4 tok/s per request and a time to first token (TTFT) of 108 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 147 users for chat use and 36 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: 1T total / 52B active, DSpark k=3, 1M context. Multimodal checkpoint measured on text. Model: mistralai/Mistral-Large-4-1T-A52B
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, Mistral-Large-4-1T-A52B (1T parameters), served in FP8 format with SGLang and speculative decoding on DGX B300 (TP=8), reached a generation speed of 157.4 tok/s per request and a time to first token (TTFT) of 108 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 147 users for chat use and 36 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.