Model benchmarks
gpt-oss:20b local LLM performance
As of April 2026, gpt-oss:20b runs at up to 76.2 tok/s for local inference (best of 27 community benchmark runs across 2 GPUs).
OllamaMXFP4
Model size
20.9B
Peak speed
76.2 tok/s
Average speed
63.5 tok/s
Min memory
n/a
Max context
9,865 tokens
Best quality
91.2
Benchmark runs
27
GPUs tested
2
Performance by hardware and tool
Every hardware/tool/quantization combination gpt-oss:20b has been benchmarked on, ranked by peak token generation speed. Last updated April 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon RX 7900 XTX | Ollama | MXFP4 | 76.2 tok/s | 65.0 tok/s | 13.1 GB | 9,865 tokens | 72.2 | 23 |
| CPU only | Ollama | MXFP4 | 75.0 tok/s | 75.0 tok/s | n/a | 8,192 tokens | 0.0 | 2 |
| NVIDIA GeForce RTX 5080 | Ollama | MXFP4 | 64.0 tok/s | 35.0 tok/s | 13.3 GB | 8,007 tokens | 61.6 | 2 |
Frequently asked questions
- How fast is gpt-oss:20b for local inference?
- Across 27 community benchmark runs, gpt-oss:20b reaches up to 76.2 tok/s and averages 63.5 tok/s, with the fastest results on AMD Radeon RX 7900 XTX.
- How much memory does gpt-oss:20b need?
- The leanest observed configuration used about n/a of memory (quantizations tested: MXFP4).
- Which tools have been used to run gpt-oss:20b?
- Benchmarks were submitted using Ollama. Results are community-contributed and updated as new runs arrive.