LLM Inference Benchmark Explorer
Gemma-4-26B-A4B-it on DGX B300 — BF16, vLLM, TP1 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | Gemma-4-26B-A4B-it |
|---|---|
| Parameters | 26B |
| Intelligence Index | 16.7 |
| Agentic Index | — |
| Device | DGX B300 |
| Quantization | BF16 |
| Max C | 64+ |
| Chat Capacity | 256+ |
| Agentic Capacity | 96+ |
| TP | — |
| DP | — |
| PP | — |
| Engine | vLLM |
| Speculative Decoding | — |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 60 | 168.20 | PASS |
| 2 | 243 | 162.99 | PASS |
| 4 | 474 | 129.25 | PASS |
| 8 | 544 | 103.98 | PASS |
| 16 | 819 | 88.97 | PASS |
| 32 | 456 | 79.91 | PASS |
| 64 | 524 | 69.33 | PASS |
In OpenZeka's measurement, Gemma-4-26B-A4B-it (26B parameters), served in BF16 format with vLLM on DGX B300, reached a generation speed of 168.2 tok/s per request and a time to first token (TTFT) of 60 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is at least 256 users for chat use and at least 96 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.
“At least” means the configuration met the speed targets even at the highest load tested, so the real capacity may be higher.
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, Gemma-4-26B-A4B-it (26B parameters), served in BF16 format with vLLM on DGX B300, reached a generation speed of 168.2 tok/s per request and a time to first token (TTFT) of 60 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is at least 256 users for chat use and at least 96 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.
“At least” means the configuration met the speed targets even at the highest load tested, so the real capacity may be higher.