RUN THIS LLM
Search local LLM hardware requirements
GPT-OSS 20B
OpenAI · 21B MoE · General
Compact OpenAI open-weight model. 21B MoE with 3.6B active params. Surprisingly strong reasoning and code for its size. 128K context. Apache 2.0.
VRAM Requirements
| Quantization | VRAM |
|---|---|
| Q4_K_M (smallest) | 12.6 GB |
| Q8_0 (balanced) | 23.1 GB |
| FP16 (full quality) | 42 GB |
Specifications
- Parameters: 21B MoE (3.6B active per token)
- Category: General
- Max context: 128K tokens
- System RAM: 16 GB minimum
- HuggingFace: openai/gpt-oss-20b
Benchmarks
- MMLU: 81 — General knowledge & reasoning (5-shot)
- HumanEval: 82 — Code generation (pass@1)
- MATH: 68 — Competition-level math reasoning
Loading interactive analysis...