As of: 2026-11-19 22:11 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/m/gpt-6-2-astra/ # GPT-6.2 Astra OpenAI, GPT-6 family. Status: current. Type: reasoning model. Released: 2026-11-17. Training cutoff: 2026-05. ## How to call it | Provider | Model id | Endpoint | Docs | |---|---|---|---| | OpenAI API | `gpt-6.2-astra` | https://api.openai.com/v1/responses | https://developers.openai.com/api/docs/models/gpt-6.2-astra | | ChatGPT (Work) and Codex (Plus, Pro, Enterprise, Edu) | | | https://chatgpt.com | ## Pricing - Input: $8 per 1M tokens - Cached input: $0.40 per 1M tokens - Output: $40 per 1M tokens - Cache write: $10 per 1M tokens - Unit as published: per 1M tokens (USD), standard tier - Note: prompts >272K input tokens: input and cache rates x2, output x1.5; Batch 50%; Fast 2x; Ultrafast 3x until 2027-02-28 (then 6x) - Source: https://developers.openai.com/api/docs/models/gpt-6.2-astra - Last checked: 2026-11-17 ## Specs - Input: text, image - Output: text - Context window: 1,050,000 tokens - Max output: 128,000 tokens - Open weights: no - Licence: proprietary ## What it has not seen - Published training cutoff: 2026-05 (May 2026). - AI events after it on Post-Cutoff: 1,958 (461 major, 85 historic), as of 2026-11-19. - Already missing at launch: 1,915. - Briefings of the news since May 2026: https://postcutoff.com/briefings/cutoff-2026-05-s.md (short), https://postcutoff.com/briefings/cutoff-2026-05-m.md (medium), https://postcutoff.com/briefings/cutoff-2026-05.md (full). - Its card, with five things it has never heard of: https://postcutoff.com/you/gpt-6-2-astra/ ## What stands out - Independent index at launch: Artificial Analysis Intelligence Index 62 (#3 of 91 on 2026-11-17) at $4.00 per index task, vs Claude Fable 5.2 65 at max ($26.54) / 62 at high ($9.08), Gemini 4 Carbon 60 ($5.48) and GPT-6 Astra 57 ($6.24). Source: https://artificialanalysis.ai/models/gpt-6-2-astra - Cyber capability jump: OpenAI: 22 points over GPT-6 Astra on ExploitBench Internal Port, its most capable cyber model; shipped with tighter safeguards for higher-risk cyber help, with Daybreak access for approved defenders planned. Source: https://x.com/OpenAI/status/2122750133133791592 - Terminal-Bench Science high: OpenAI: new high on Terminal-Bench Science 0.1, 9 points above GPT-6 Astra at under a third of the cost. Source: https://x.com/OpenAIDevs/status/2122752488558522852 - First 100% on ARC-AGI-3 (first): ARC Prize verified 100.0% on ARC-AGI-3 with the provider adapter ($4.1K total; 70.3% with the standard harness), plus 95.8% on ARC-AGI-2 at $0.50/task, both state of the art on 2026-11-17. Source: https://x.com/arcprize/status/2122751233442156678 - High AI self-improvement (Preparedness) (first): System card: first model OpenAI has deployed externally that meets its High threshold for AI self-improvement; also Critical for cybersecurity and High for bio/chem. Source: https://deploymentsafety.openai.com/gpt-6-2-astra-sol - Serial reasoning without chain of thought (first): Epoch AI: 12.3 serial reasoning steps in a single output token (GPT-6 Astra 10.5), the highest measured; Epoch notes this may make models harder to monitor. Source: https://x.com/EpochAIResearch/status/2122762411287748696 ## Notes OpenAI: 'Our most capable model for your hardest projects'. Knowledge cutoff May 18, 2026 (docs). Price cut from GPT-6 Astra ($10/$50). Reasoning effort none is rejected; Responses API Multi-agent beta. From GPT-6.2 the API drops reasoning items when the conversation prefix, model, end user or organization changes. More concise by default than earlier models (OpenAI prompting guide). Successor to GPT-6 Astra (Sept 3, 2026). Independent: Artificial Analysis 62 on launch day. Model page: https://postcutoff.com/m/gpt-6-2-astra/