GPT-6.2 Astra
OpenAI Released Nov 17, 2026 Proprietary 1.05M context
Last verified: Nov 18, 2026
GPT-6.2 Astra is OpenAI's new frontier model for the hardest projects, released November 17, 2026 alongside the efficiency-focused GPT-6.2 Sol. It prices 20% below GPT-6 Astra while scoring higher on agentic benchmarks.
What changed
- OpenAI's new frontier model for your hardest projects — available starting today in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu, with API model ID gpt-6.2-astra.
- Priced $8 input / $40 output per 1M tokens — 20% below GPT-6 Astra — with a $0.40 cached-input rate (60% discount).
- First externally released OpenAI model to meet the High threshold for AI self-improvement under the Preparedness Framework.
Background
- GPT-6 Astra began rolling out to select customers around July 2026 with enhanced cybersecurity safety protocols.
- GPT-6 Cyber became available roughly two weeks before this release (early November 2026).
- Approved defenders still use GPT-6 Astra and GPT-6 Cyber via OpenAI's Daybreak program; OpenAI says it will work to bring GPT-6.2 models to Daybreak.
Astra or Sol? — pick the right variant
Where you can use GPT-6.2 Astra
| API | Model ID gpt-6.2-astra — live now; Ultrafast available at launch. |
|---|---|
| ChatGPT Work & Codex | Plus, Pro, Enterprise and Edu — available starting today. |
| Ultrafast | Pro 500 subscribers (Pro 200 gets Ultrafast first with GPT-6.2 Sol). |
| Enterprise / Edu | Requires workspace admin enablement. |
How GPT-6.2 Astra compares
| GPT-6.2 Astra | |
|---|---|
| Release date | Nov 17, 2026 |
| Context window | 1.05M |
| Parameters | Undisclosed |
| Key benchmark | GPQA Diamond 97% |
| API input | $8/M |
Benchmarks
GPT-6.2 family — official benchmark board
As of Nov 17, 2026 · Scores transcribed from OpenAI's GPT-6.2 model-card tables (via DataCamp)
| Benchmark | GPT-6.2 Astra | GPT-6.2 Sol | GPT-6 Astra | GPT-6.1 Sol | Claude Fable 5.2 |
|---|---|---|---|---|---|
| Agents' Last Exam v1.1 · cost per task | 78.3% | 73.0% | 73.1% | 72.5% | 72.8% |
| GDPval v2 (OpenAI internal) | 50.4% | 46.9% | 44.2% | 41.5% | 44.8% |
| Harvey LAB-AA v1.1 · new state of the art | 15.0% | 11.1% | 8.6% | 6.9% | 7.2% |
| AutomationBench 1.0.6 · Gemini 4 Carbon leads at 53.9% | 52.4% | 41.2% | 41.4% | 36.1% | 51.4% |
| LifeVal | 86.1% | 83.8% | 79.8% | 79.5% | — |
| FrontierCode 2.0 (Cognition) · Claude Opus 5.5: 45.5% at max effort | 44.2% | 43.8% | 42.8% | 39.0% | 49.1% |
| SecureSWE (OpenAI internal) | 65.7% | 55.6% | 48.7% | 49.7% | — |
| PatchBench | 79.7% | 75.6% | 70.9% | 71.0% | — |
| ExploitBench Internal Port · cyber benchmarks run with safeguards disabled | 53.5% | 50.5% | 31.5% | 21.5% | — |
| Terminal-Bench-Science 0.1 | 77.1% | 68.0% | 68.1% | 57.0% | 73.3% |
| Vals Science Index v1 | 68.4% | 58.5% | 65.1% | 59.6% | 63.7% |
| BenchCAD 2.1 Core | 82.5% | 73.1% | 62.6% | 52.6% | 83.5% |
| AA Intelligence Index v5.0.1 | 62.1 | 57.3 | 56.9 | 54.6 | 64.6 |
| GPQA Diamond (no tools) | 97.0% | 96.0% | 96.0% | — | — |
| AA-3D (Elo) · Astra ranks first | 1318 | 1259 | — | — | 1271 |
| AA-AnalystAgent · correct on all five independent attempts — new high | 62.5% | — | — | — | — |
API pricing
Example: 1M input + 1M output tokens = $48.
Source: https://platform.openai.com/pricing. Prices last checked Nov 18, 2026.
How to use GPT-6.2 Astra
- Sign up on the OpenAI platform to get an API key.
- Pricing & console: https://platform.openai.com/pricing
curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{"model": "gpt-6.2-astra", "input": "Hello"}'API behavior
- The API rejects requests that set reasoning effort to none.
- Responses API supports the Multi-agent beta for delegating work to subagents.
- Discounted background completion windows are in public beta.
- Five reasoning efforts from low to max — Sol at high effort scores 72.0% on Agents' Last Exam at $0.69/task vs GPT-6 Astra max effort at $8.99/task.
- Prompt-caching guidance from OpenAI: keep prompt prefixes unchanged and append new instructions.
What GPT-6.2 Astra can do
- Professional outputs: slides, spreadsheets and documents that follow field conventions (law, sales, finance, accounting) and match templates down to exact font sizes.
- Computer use precise enough for CAD and simulation — OpenAI demos include car-chassis dynamics (3D model + Python solver + Blender animation), an RC buggy and an inline-six engine.
- Web and 3D design: a playable pinball table, an explorable Colosseum and a 'Meridian Sound Racing' game built from prompts.
- Coding: cleaner style, more focused tests, fewer revision rounds and better tool/effort judgment.
- Cybersecurity: finds and exploits vulnerabilities more effectively, writes more secure code and produces better patches — behind safeguards for high-risk assistance.
Safety & Preparedness
- Cybersecurity: Critical threshold (Preparedness Framework)
- Biological / chemical: High threshold
- AI self-improvement: High — the first externally released OpenAI model at this level
Mitigations
- Tighter safeguards for high-risk cybersecurity assistance — they restrict some legitimate vulnerability-research and security-testing workflows.
- Strengthened containment plus Control Room, which can halt model output.
- Expanded monitoring can flag and pause unauthorized behavior.
System card: https://deploymentsafety.openai.com
Third-party signals
GPT-6.2 Astra FAQ
What is GPT-6.2 Astra?
OpenAI's frontier model for the hardest projects, released November 17, 2026 alongside GPT-6.2 Sol. API model ID: gpt-6.2-astra.
How much does GPT-6.2 Astra cost?
$8 per 1M input tokens and $40 per 1M output tokens, with cached input at $0.40 — 20% below GPT-6 Astra on input/output and a 60% cached-input discount.
What is GPT-6.2 Astra's context window?
1,050,000 tokens total: 922,000 max input and 128,000 max output. Knowledge cutoff: May 18, 2026.
Where can I use GPT-6.2 Astra?
In ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu plans, and via the API with model ID gpt-6.2-astra. Ultrafast is available to Pro 500 subscribers.
How does GPT-6.2 Astra compare to GPT-6 Astra?
It scores 5.2 points higher on Agents' Last Exam at under one-third the cost per task ($2.89 vs $8.99), with 20% lower token prices and a 60% cached-input discount.
How does GPT-6.2 Astra compare to Claude Fable 5.2?
GPT-6.2 Astra is 5.5 points ahead on Agents' Last Exam at about 5% of the cost per task. Claude Fable 5.2 still leads FrontierCode 2.0 (49.1% vs 44.2%), BenchCAD (83.5% vs 82.5%) and the AA Intelligence Index (64.6 vs 62.1).
Is GPT-6.2 Astra safe to deploy?
It meets the Critical threshold for cybersecurity and the High threshold for biological/chemical capabilities under OpenAI's Preparedness Framework, and is the first externally released OpenAI model at the High threshold for AI self-improvement — shipped with tightened safeguards, Control Room output halting and expanded monitoring.