gpt-oss-120b — Pantheon
  
- 

  
  
  
  
  
  
  
  
  
  
  
  
- 
  
  
  

  
    
      [← Pantheon](../)
      [copy as markdown](index.md)
    

    # gpt-oss-120b

    
OpenAI · open weights released 5 Aug 2025 · Apache 2.0
    
The larger of the two open-weight models OpenAI released on 5 August 2025 — its first open weights since [GPT-2](../gpt-2/) in 2019. A ~117B-parameter mixture-of-experts reasoner (~5.1B active) sized to run on a single 80GB GPU, released alongside the smaller [gpt-oss-20b](../gpt-oss-20b/).
    
Thin page — the launch facts are here; a corpus/dossier pass is pending (the dbs cover Aug 2025, so the tweet layer is minable). Sibling: [gpt-oss-20b](../gpt-oss-20b/).

    
## Sources

    
### Official

    

      
- 2025-08-05 [Introducing gpt-oss](https://openai.com/index/introducing-gpt-oss/) — the announcement: both sizes, Apache 2.0, the first-open-weights-since-GPT-2 framing.
      
- 2025-08-05 [openai/gpt-oss-120b](https://huggingface.co/openai/gpt-oss-120b) (Hugging Face) — the weights.
      
- tk — model card / paper link; mirror the announcement + card.
    
    
### Writing & commentary

    
tk — Zvi and day-of coverage to collect at the dossier pass.
    
### Tweets

    
tk — corpus pass pending. Records will populate when cited.

    
## Official record

    

      
- Released 5 August 2025, open weights, Apache 2.0 — OpenAI’s first open-weights release since GPT-2 (Nov 2019).
      
- As published: ~117B-parameter MoE, ~5.1B active per token; positioned as o4-mini-class on reasoning benchmarks; fits a single 80GB GPU; adjustable reasoning effort; the “harmony” response format.
      
- tk — benchmark table as published, exact context window, safety-evaluation framing (the announcement discussed worst-case-fine-tuning testing).
    

    
## History

    

      
- 2025-08-05 World at release: two days before [GPT-5](../gpt-5/)’s launch; the open-weights field it entered was led from China ([DeepSeek-R1](../deepseek-r1/), [Kimi K2](../kimi-k2/), [Qwen 3](../qwen-3/), [GLM-4.5](../glm-4-5/)) — the release was widely read against that field and against OpenAI’s own six-year closed run since GPT-2.
      
- tk — reception arcs (benchmark-vs-vibes discourse, the hallucination-rate discussion), deployment story, whether it develops a distinct character from its small sibling.
    

    
## Impressions

    
tk — corpus pass pending; no attributed reads collected yet.

    
    
    
[← back to the Pantheon](../)