Model benchmarks
gemma4:12b-mlx local LLM performance
As of August 2026, gemma4:12b-mlx runs at up to 68.5 tok/s for local inference (best of 2 community benchmark runs across 2 GPUs).
Ollamanvfp4
Model size
12.4B
Peak speed
68.5 tok/s
Average speed
57.8 tok/s
Min memory
9.3 GB
Max context
5,529 tokens
Best quality
85.6
Benchmark runs
2
GPUs tested
2
Performance by hardware and tool
Every hardware/tool/quantization combination gemma4:12b-mlx has been benchmarked on, ranked by peak token generation speed. Last updated August 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| Apple M5 Max | Ollama | nvfp4 | 68.5 tok/s | 68.5 tok/s | 9.3 GB | 4,792 tokens | 60.6 | 1 |
| Apple M4 Pro | Ollama | nvfp4 | 47.1 tok/s | 47.1 tok/s | 9.3 GB | 5,529 tokens | 75.1 | 1 |
Frequently asked questions
- How fast is gemma4:12b-mlx for local inference?
- Across 2 community benchmark runs, gemma4:12b-mlx reaches up to 68.5 tok/s and averages 57.8 tok/s, with the fastest results on Apple M5 Max.
- How much memory does gemma4:12b-mlx need?
- The leanest observed configuration used about 9.3 GB of memory (quantizations tested: nvfp4).
- Which tools have been used to run gemma4:12b-mlx?
- Benchmarks were submitted using Ollama. Results are community-contributed and updated as new runs arrive.