Model benchmarks
gemma4-26b-a4b-qat local LLM performance
As of October 2026, gemma4-26b-a4b-qat runs at up to 65.5 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
Model size
26B
Peak speed
65.5 tok/s
Average speed
62.2 tok/s
Avg PP
824.0 tok/s
Min memory
20.9 GB
Max context
65,536 tokens
Avg output / run
10,132 tokens
Avg runtime / run
3m 8s
Avg quality
69.7
Benchmark runs
2
GPUs tested
1
Quality by task
Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 2 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.
| Task | P5 (low) | Avg | P95 (high) |
|---|---|---|---|
| Overall | 67.7 | 69.7 | 71.7 |
| Agent Workflow | 58.6 | 68.9 | 79.3 |
| Code Generation | 66.9 | 69.7 | 72.5 |
| Role Play & Narrative | 74.0 | 78.6 | 83.2 |
| Research & Analysis | 61.1 | 61.5 | 61.9 |
Performance by hardware and tool
Every hardware/tool/quantization combination gemma4-26b-a4b-qat has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| NVIDIA RTX A2000 12GB | llama.cpp | — | 65.5 tok/s | 62.2 tok/s | 20.9 GB | 65,536 tokens | 69.7 | 2 |
Benchmark runs
All 2 gemma4-26b-a4b-qat runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is gemma4-26b-a4b-qat good for coding?
- In our benchmarks, gemma4-26b-a4b-qat scores 69.7/100 for coding. It runs at about 62.2 tok/s, so if you want more speed, Nex-N2.5-mini-IQ3_XXS is faster (~91.4 tok/s) and still scores well for coding (81.6/100).
- Is gemma4-26b-a4b-qat good for agentic (tool-using) tasks?
- In our benchmarks, gemma4-26b-a4b-qat scores 68.9/100 for agentic workflows. It runs at about 62.2 tok/s, so if you want more speed, Nex-N2.5-mini-APEX-Mini is faster (~88.2 tok/s) and still scores well for agentic workflows (91.3/100).
- How fast is gemma4-26b-a4b-qat for local inference?
- Across 2 community benchmark runs, gemma4-26b-a4b-qat reaches up to 65.5 tok/s and averages 62.2 tok/s, with the fastest results on NVIDIA RTX A2000 12GB.
- How much memory does gemma4-26b-a4b-qat need?
- The leanest observed configuration used about 20.9 GB of memory.
- Which tools have been used to run gemma4-26b-a4b-qat?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.