OpenAI: gpt-oss-20b
Made by openai. Open weights
- Input price
- $0.018
- per 1M tokens
- Output price
- $0.09
- per 1M tokens
- Context window
- 131K
- tokens
- Quality
- 1,288
- #257 on LMArena
- Size
- 21B
- parameters
- Downloads
- 6.1M
- on Hugging Face, last 30 days
Can I run it locally?
Estimate for the Q4_K_M version with an 8K context. Pick your own card and settings in the calculator.
- NVIDIA GeForce RTX 40608 GBRuns, but slowly5.2 to 7.3 tokens/s
- NVIDIA GeForce RTX 306012 GBRuns, but slowly10 to 15 tokens/s
- NVIDIA GeForce RTX 4060 Ti 16 GB16 GBRuns14 to 19 tokens/s
- NVIDIA GeForce RTX 409024 GBRuns47 to 67 tokens/s
- NVIDIA GeForce RTX 509032 GBRuns84 to 119 tokens/s
- Apple M1 Pro 32 GB32 GBRuns9.4 to 13 tokens/s
Prices: OpenRouter, refreshed 8 Oct 2026. Quality: LMArena (CC BY 4.0), leaderboard of 2 Oct 2026. Model page on Hugging Face