GPT-OSS (20B, MoE)
OpenAI's open-weight 20B-parameter Mixture-of-Experts model (~3.6B active per token). Configurable reasoning effort (low/medium/high), full chain-of-thought access, and native function-calling. Lower-latency frontier-class on-device option for machines with 16GB+ memory.
At a glance
- Parameters
- 20B
- Quantization
- Q4_K_M
- Context window
- 131K tokens
- Approx. size
- 12.0 GB
- Engines
- llama.cpp, MLX, Ollama
- License
- Apache-2.0 (open)
- Version
- v1.1.0 , released 2026-04-26
- Family
- gpt-oss
Tuning
Defaults a gezel client applies out of the box, from real evaluation runs against this exact quantization.
- reasoning.effort
"medium"- reasoning.thinkingBudget
6144
Model behaviors
Client-side behavior modules the catalog enables for this model (reasoning-tag handling, fabrication detection, prompt shaping).
- reasoning.strip-channel-tags
- prompt.private-reasoning-guidance
- prompt.prefer-writefile-edits
- prompt.derive-by-execution
- fabrication.detect-past-tense-no-tools
- turn.preamble-folding
Sources
- llama.cpp (GGUF)
- bartowski/openai_gpt-oss-20b-GGUF ·
openai_gpt-oss-20b-Q4_K_M.gguf· Q4_K_M - MLX
- mlx-community/gpt-oss-20b-MXFP4-Q4
- Ollama
gpt-oss:20b- Upstream
- https://ollama.com/library/gpt-oss
- Manifest
- View on GitHub
Tags
- openai
- tools
- reasoning
- moe
- open-weights