GPT OSS 20B
openai/gpt-oss-20b- OpenAI
- Family · GPT OSS
- Released 5 Aug 2025
- Open weights
- Text → Text
- Reasoning · adjustable
- Tools · JSON
Spec & list price
Context128K
Max output118K
Input / 1M$0.03
Output / 1M$0.134.3× input
Scores · standing
Intelligence10.0#245/489
Coding20.7#142/196
Usage
Tokens / day20.7B▼ 2.6% 7d
Usage rank#71
Requests / day3.9M
Tool calls / req0.07
Per request · 30 days
Avg input3.2K0.2× median
Avg output6200.9× median
Cache hit18%0.3× median
Reasoning share70%1.2× median
Tool-call failures3.1%1.7× median
Consumption
Peak 49.6B 2 OctLow 15.4B 1 NovRank today #71
News & trends
5 mentions this weekthis model · videos, posts, articles · 30 days
Trending · 14 days
- Pi + gpt-oss 20B Added Type Hints Without Breaking Tests #shorts665 views
- I Built a Minecraft Style Game Using Qwen 3.8 27B and GPT OSS 20B on RX 9060XT31 views
News · latest from OpenAI
- GPT-6.1 Terra vs. Claude Opus 5.5: Cheaper and faster in three tests at maxthenewstack.io · 2 Nov
- Chinese labs now target OpenAI's GPT-6 Astra for distillationfork.zone · 1 Nov
- GPT-6.1 Terrasimonwillison.net · 31 Oct
- GPT-5.6 Sol can beat Factorio on peaceful moderyanmadden.net · 31 Oct
Benchmarks
bar = share of the top score · Benchmark data: Artificial AnalysisSpeed · median across 11 providers, latest day
Output speed#38/334132 tok/s2.0× faster than median
Fastest provider450 tok/sbest of 11 providers
First token#310.4 s2.5× quicker than median
First answer token#4210.0 sthinking included
Best value in its score band
Value
Intelligence Index
Blended price per 1M tokens (log)
Family progress
Intelligence Index
- Family releases10.0
Technical details
- Reasoning
- on · medium
- Tokenizer
- GPT
- Knowledge cutoff
- 2024-06-30
- License
- Apache 2.0
- Parameters
- 20.9B
- Moderation
- none
- Downloads · likes
- 6.0M · 5.3K
Reasoning effort
- high
- medium · default
- low
Request parameters 20
- tools
- tool_choice
- response_format
- structured_outputs
- reasoning
- reasoning_effort
- include_reasoning
- temperature
- top_p
- top_k
+10 more
- min_p
- max_tokens
- stop
- seed
- frequency_penalty
- logit_bias
- logprobs
- presence_penalty
- repetition_penalty
- top_logprobs
When to use
Best for
- Low-cost self-hosted reasoning
- Math and code help on modest hardware
- Structured extraction and tool calling
- Fine-tuning an open model for a niche domain
Think twice for
- Image or audio input
- Hardest science and agentic tasks
- Very long document analysis
Research papers
- Pre-Crime for Agents, at the Last Second: Probing When a Recruited Agent Gives In29 Oct
- Heads of Security: Learned Attention Gates Improve Prompt Injection Robustness27 Oct
- How Much Does the Hint Matter? Scope Conditions for Hindsight Self-Distillation in Financial Reasoning26 Oct
- Training an MoE LLM in MXFP4: How We Closed the Gap to BF1623 Oct
- Diagnosing the Explicit-to-Applied Theory of Mind Gap in LLMs22 Oct
- MuninAI: A Self-Hosted Agentic Framework for Independent Research Groups16 Oct
- Towards Robust Calibration Dataset for Domain-General Mixture of Experts Pruning11 Oct
- RouterInterp: Understanding Superposed Specialisation in Mixture of Experts Routing8 Oct
Social
Most watched and discussed · 90 days
- Pi + gpt-oss 20B Added Type Hints Without Breaking Tests #shortsMishka · 665 views · 29 Oct
- I Built a Minecraft Style Game Using Qwen 3.8 27B and GPT OSS 20B on RX 9060XTUnknown Programmer · 31 views · 22 Oct
- 【検証】SpaceQA-gpt-oss-20bにコーディング全52問を投げた結果、リトライデコレータで誤答… #Shorts #LLM電霞リリ · 2 views · 26 Oct
- 【公開実験】SpaceQA-gpt-oss-20b ベンチマーク｜Future Craft Lab電霞リリ · 2 views · 25 Oct
- 【意外性】SpaceQA-gpt-oss-20b、日英翻訳のニュアンスで見えた意外な弱点 #Shorts #LLM電霞リリ · 1 views · 25 Oct
Same family
GPT OSS- 11.6
Alternatives
other vendors · similar capability- 12.9
- 12.3
- 14.9
- 6.7
Using GPT OSS 20B at volume?
Sign in to ask for a volume discount. You will continue in the OSMEXO dashboard.