GPT-OSS Models — Hardware Requirements

10 GPT-OSS models from OpenAI and the community, from the smallest that runs in 0.3 GB of VRAM up to 120B parameters. Every row links to full quantization tables and GPU compatibility.

All GPT-OSS Models by Size

ModelParamsContext
GPT OSS 20B NPU220B131K
GPT OSS 20B20.9B131K
Huihui GPT OSS 20B BF16 Abliterated20.9B131K
GPT OSS 20B Heretic20.9B131K
GPT OSS 20B RichardErkhov Heresy21.5B131K
GPT OSS Safeguard 20B21.5B131K
GPT OSS 20B Heretic Ara v321.5B131K
GPT OSS 120B116.8B131K
GPT OSS 120B Eagle3 v3120B131K

How GPT-OSS Compares — Benchmark Rating

GPT OSS 120B is the highest-rated GPT-OSS model with an overall benchmark rating of 58.0/100 — #12 among 78 open models. The top proprietary model, Claude Fable 5 (high), scores 87.8. Click a model to see its full benchmark breakdown.

Claude Fable 5 (high) · proprietary87.8
Claude Opus 4.7 · proprietary86.4
Claude Fable 5 Max · proprietary82.4
GPT 6 Astra (high) · proprietary82.4
GPT 6 Astra Max · proprietary82.4
GLM 565.9
Composite of normalized public benchmark scores (methodology) · ■ GPT-OSS · ■ other models

Frequently Asked Questions

How much VRAM do I need to run a GPT-OSS model?
The smallest GPT-OSS model, GPT OSS 20B RichardErkhov Heresy, runs from 6.3 GB of VRAM at an aggressive quantization. Larger family members need proportionally more — see the table above for every model.
Which GPT-OSS models can I run on a 16 GB GPU?
7 of 9 GPT-OSS models fit in 16 GB of VRAM at some quantization, including GPT OSS 20B, Huihui GPT OSS 20B BF16 Abliterated, GPT OSS 20B RichardErkhov Heresy.
What is the most popular GPT-OSS model to run locally?
GPT OSS 20B is the most downloaded GPT-OSS model in local-friendly quantized formats. It runs from 9.3 GB of VRAM.
How do GPT-OSS models score on benchmarks?
GPT OSS 120B leads the family with an overall benchmark rating of 58.0/100, ranking #12 among 78 open models, while the top proprietary model, Claude Fable 5 (high), scores 87.8. See the comparison chart above for the full standings.