Introducing GPT-5.2

• GPT-5.2 released Dec 11, 2025; three variants: Instant, Thinking, Pro.
  • 400K context window, 128K output, August 2025 knowledge cutoff.
• Beats industry pros on 70.9% of GDPval knowledge-work comparisons.
  • Spans 44 occupations: code, spreadsheets, presentations, analysis.
• Thinking: 92.4% GPQA Diamond, 100% AIME 2025 no-tools.
• Pro: 93.2% GPQA Diamond, 80% SWE-Bench Verified, fewest major errors.
• Instant: fast everyday workhorse — writing, translation, how-tos.

X discourse

• @OpenAIDevs: “We’re retiring older models in Codex: gpt-5.2-codex, gpt-5.1-codex-mini, others.” (4676 likes)
• @pankajkumar_dev: “GPT-5.5 ‘Spud’: new foundation, big model smell, omnimodal agentic system.” (1486 likes)
• @LukasHozda: “GPT-5.5 (Spud) is a new pre-trained base… How can they be this bad at versioning.” (908 likes)
• @chatgpt21: “GPT 5.4 w/codex & Opus 4.6 have made a sizable dent in automating tedious labor.” (488 likes)
• @emollick: “as soon as models crossed threshold (GPT-5.2), suddenly Claude Code & Codex viable for agents.” (133 likes)
• @ArtificialAnlys: “MiniMax-M2.7 equivalent to GLM-5… ahead of GPT-5.2 on agentic tasks.” (131 likes)

· ** · Read on openai.com