gpt-oss-20b
gpt-oss-20b is an open-weights language model from OpenAI, released August 5, 2025. It has 21B parameters (mixture-of-experts, 4.2B active). The context window is 128K tokens. On the Epoch Capabilities Index it scores 135.0, rank 143 of 256 rated models.
Check it against your own PC →
Can you run gpt-oss-20b locally?
The smallest common GPU it runs on is 6 GB of VRAM (with 32 GB of RAM): Q6_K, GPU + experts in RAM, roughly 14 tokens per second. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Runs · GPU + CPU experts | Q6_K (≈ lossless) | 5.2 GB | 13.0 GB | ~14 |
| 8 GB VRAM | Runs · GPU + CPU experts | Q6_K (≈ lossless) | 7.2 GB | 11.0 GB | ~16 |
| 12 GB VRAM | Runs · GPU + CPU experts | Q6_K (≈ lossless) | 11.2 GB | 7.0 GB | ~23 |
| 16 GB VRAM | Runs on GPU | Q4_K_M (very good) | 13.8 GB | – | ~100 |
| 24 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 18.2 GB | – | ~140 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 18.2 GB | – | ~240 |
Ready-made quantizations on Hugging Face: unsloth/gpt-oss-20b-GGUF · ggml-org/gpt-oss-20b-GGUF · lmstudio-community/gpt-oss-20b-GGUF · bartowski/openai_gpt-oss-20b-GGUF · mlx-community/gpt-oss-20b-MXFP4-Q8 · lmstudio-community/gpt-oss-20b-MLX-8bit
Benchmark results
| Benchmark (Epoch AI) | Score |
|---|---|
| OTIS Mock AIME | 65.2% |
| GPQA Diamond | 49.7% |
| Terminal-Bench | 3.4% |
Links
Announcement · Hugging Face: openai/gpt-oss-20b · Epoch AI benchmarks
Models with a similar score
DeepSeek-R1 · Qwen3-235B-A22B · Qwen3-30B-A3B-Thinking (Jul 2025) · Qwen3-235B-A22B-Instruct (Jul 2025) · Grok 3 · Qwen3-32B · QwQ-32B · GPT-4.5
More from OpenAI
GPT-6.1 Terra · GPT-6.1 Sol · GPT-6 Sol · GPT-6 Luna · GPT-6 Astra · GPT-5.6 Cyber · Daybreak Red · Daybreak Blue