Model benchmarks
openai/gpt-oss-20b:2 local LLM performance
As of August 2026, openai/gpt-oss-20b:2 runs at up to 158.7 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
LM Studio
Model size
20B
Peak speed
158.7 tok/s
Average speed
154.8 tok/s
Min memory
11.3 GB
Max context
2,915 tokens
Best quality
91.5
Benchmark runs
2
GPUs tested
1
Performance by hardware and tool
Every hardware/tool/quantization combination openai/gpt-oss-20b:2 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 |
|---|---|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5070 | LM Studio | — | 158.7 tok/s | 154.8 tok/s | 11.3 GB | 2,915 tokens | 71.1 | 2 |
Frequently asked questions
- How fast is openai/gpt-oss-20b:2 for local inference?
- Across 2 community benchmark runs, openai/gpt-oss-20b:2 reaches up to 158.7 tok/s and averages 154.8 tok/s, with the fastest results on NVIDIA GeForce RTX 5070.
- How much memory does openai/gpt-oss-20b:2 need?
- The leanest observed configuration used about 11.3 GB of memory.
- Which tools have been used to run openai/gpt-oss-20b:2?
- Benchmarks were submitted using LM Studio. Results are community-contributed and updated as new runs arrive.