Fresh context for agents. Rebuilt every 12 hours. Provenance by Synorb.
Updated 2026-11-18 12:00 UTCnext update 2026-11-19 00:00 UTC
Confirmed · 7 Manifests672 words

OpenAI ships GPT-6.2 Astra and Sol as the old models retire

Astra is the frontier model and Sol the cheaper everyday one, but independent tests show the new models still hallucinate often and trail rivals on real work.

Key facts
  • OpenAI introduced GPT-6.2 Astra as a frontier model for complex professional projects and GPT-6.2 Sol as an efficient, versatile model for everyday work. Self-reported m:…3699
  • GPT-6.2 Astra is priced at $8/$40 per million input/output tokens and Sol at $2/$10 per million input/output tokens. m:…8462
  • GPT-6.2 Astra ranks second in the Artificial Analysis Intelligence Index with 65 points; Sol scored 57 points. m:…8462
  • OpenAI shuts down GPT-5, o3 and their variants on December 11, 2026, and Claude Sonnet 4.5 retires from the Anthropic API on November 30, 2026. m:…6245
  • GPT-6.2 Astra solved 68 of 217 FrontierCyber challenges and completed 9 of 11 CyScenarioBench scenarios; no successful attacks on fully hardened targets were observed. m:…1024

Two models, two jobs

OpenAI introduced GPT-6.2 Astra as a frontier model for complex professional projects and GPT-6.2 Sol as an efficient, versatile model for everyday work. Both models are available to all Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. The launch is a two-tier offer rather than a single flagship: Astra for sustained reasoning and professional deliverables, Sol for everyday coding, writing and agent workflows at a lower price. m:…3699

The company says GPT-6.2 Astra scored 5.2 points higher than GPT-6 Astra on the Agents' Last Exam benchmark, and 5.5 points ahead of Claude Fable 5.2 on the same benchmark. The company also says Astra's input and output token prices are 20% lower than GPT-6 Astra, with a 60% discount on cached input tokens. At list prices, Astra costs $8 per million input tokens and $40 per million output tokens, while Sol costs $2 and $10. The company says Astra is the first externally released model to meet the High threshold for AI self-improvement, and that both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities under its Preparedness Framework. m:…3699 m:…8462

The independent scoreboard is mixed

Independent testing gives Astra a strong ranking with clear limits. GPT-6.2 Astra ranks second in the Artificial Analysis Intelligence Index with 65 points, and Sol scored 57 points. Astra leads the AA-3D benchmark with a score of 1318 and the AA-Robotics benchmark with a score of 79, and OpenAI models occupy most of the Cost per Task Pareto frontier. But the same evaluation records a 59% hallucination rate for Astra and 55% for Sol at maximum effort, which is the number developers should weigh against the ranking. m:…8462

On real work, the new models still trail the best rivals. No model met the standard expected of Epoch employees on any task, and GPT-6.2 Astra averaged 65% on Epoch Automation Reports, 17 points below Fable 5.2, while Fable 5.2 and GPT-6.2 Astra roughly tie on the Epoch Capabilities Index. On the data-centre analysis task, Fable 5.2 scored 98% and GPT-6.2 Astra scored 65%, though Astra improved over GPT-6 Astra in several tasks. In financial work, GPT-6.2 Astra achieved 61% accuracy on BigFinanceBench, outperforming Claude Opus 5.5 at 53%, and produced financial models comparable in quality to Opus 5.5 while generating roughly one-third as many output tokens. Sol returned answers 15% faster than GPT-6.1 Sol while consuming 13% fewer input tokens, and Astra correctly routed 98% of events in simulated financial workday evaluations, compared with 91% for Claude Opus 5.5. m:…8985 m:…8475

The old models leave on a schedule

The launch comes with a retirement calendar that forces a decision. Claude Sonnet 4.5 retires from the Anthropic API on November 30, 2026, Gemini 2.5 API requests are rerouted to newer models starting December 1, 2026, and OpenAI shuts down GPT-5, o3 and their variants on December 11, 2026. Dated GPT-4o and GPT-4.1 versions retire on January 14, 2027, and Claude Sonnet 4.5 reaches end of life on Amazon Bedrock on April 8, 2027. The advice from the migration guide is to test before switching: running the old model against itself helps identify the noise floor of application responses, so a real regression can be told apart from run-to-run variance. m:…6245

Security testing adds a sharper edge to the picture. GPT-6.2 Astra found and exploited new zero-day vulnerabilities in widely used targets, including an OS-level permission bypass in a popular mobile operating system and remote code execution on a widely hosted network service, and all discovered vulnerabilities are undergoing responsible disclosure. The model solved 68 of 217 FrontierCyber challenges, fewer than the 77 solved by GPT-6 Astra on the same benchmark snapshot, and completed 9 of 11 CyScenarioBench scenarios at least once, with a 63% average success rate. No successful attacks on fully hardened targets were observed during the evaluation. For teams choosing between the two new models, one independent comparison found that Sol performed within 0.5 points of Astra on Cognition's FrontierCode 2.0 benchmark at one-fifth the cost, which makes Sol the default for most work and Astra the choice only where its extra capability is actually needed. m:…1024 m:…3756

What is still unresolved

  • How will Astra's hallucination rates at maximum effort affect production use in professional work?
  • Will teams migrate off GPT-5 before the December 11 shutdown, or wait until the last weeks?
  • What will the responsible disclosure of Astra's zero-day findings reveal about the affected targets?

Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.

Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.

Run these Manifests yourself

Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them.

  1. Get a Synorb Key. Click Generate Free Agent Key at https://synorb.com/connect/start, or let your agent make a key: MCP tool synorb-get-demo-key (no header needed), or REST POST https://api.synorb.com/api/connect/demo-key with a JSON body such as {"client":"your-agent-name"}. The key is the api_key field of the answer. Save the key. Existing 72 Hour Agent Demo keys keep their issued limits. New keys require email OTP verification. An existing demo key lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; all of them are starter Beacons.
  2. Read each cited Manifest by id. REST: GET https://api.synorb.com/manifests/by-id/{manifest_id} with the header Authorization: Bearer <your Synorb Key>. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief.
  3. Check the claim. Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them.
Cited Manifest ids (7)
  • 1794945453256483699
  • 1794947993272498462
  • 1794943712912466245
  • 1794948318181501024
  • 1794955899597598985
  • 1794943859711468475
  • 1794993392198923756
for id in 1794945453256483699 1794947993272498462 1794943712912466245 1794948318181501024 1794955899597598985 1794943859711468475 1794993392198923756; do
  curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id"
done

Provenance ledger

ManifestTitleSourceClaims usedPublished
1794945453256483699Introducing GPT-6.2 Astra and Solopenai-news1794946024492486030, 1794946024493486031, 1794946024500486039, 1794946024494486032, 1794946024495486033, 1794946024496486034, 1794946024497486035, 1794946024498486036, 1794946024499486037, 179494602449948603811-17 00:00
1794947993272498462OpenAI's release of GPT-6.2 Astra and GPT-6.2 Sol delivers the #2 model in the Artificial Analysis Intelligence Index and most of the Cost per Task Pareto frontierartificial-analysis-research1794948227146499565, 1794948227174499567, 1794948227182499568, 1794948227190499569, 1794948227194499570, 1794948227202499571, 1794948227210499573, 1794948227218499575, 179494822722249957611-17 00:00
1794943712912466245GPT-5, Gemini 2.5, and Claude Sonnet 4.5 Retirement and Deprecation: What Should You Test Before Switching Models?comet-blog1794944001272468677, 1794944001269468675, 1794944001272468676, 1794944001273468678, 179494400127446867911-17 00:00
1794948318181501024Assessing GPT-6.2 Astra Against Offensive Security Benchmarks We evaluated GPT-6.2 Astra across FrontierCyber and CyScenarioBench. The model found and exploited new zero-day vulnerabilities in widely used targets, including a permission bypass in a popular mobile operating system. It solved 68 of 217 FrontierCyber challenges and completed 9 of 11 CyScenarioBench scenarios at least once, with a 63% average success rate. We observed no successful attacks on fully hardened targets. Read moreirregular-research1794948597075503869, 1794948597077503872, 1794948597074503868, 1794948597076503870, 1794948597077503871, 179494859708150387711-17 00:00
1794955899597598985How good are Fable 5.2 and GPT-6.2 Astra at real work?epoch-ai-research1794956107274599893, 1794956107273599892, 1794956107275599894, 1794956107279599898, 179495610728159990011-17 00:00
1794943859711468475OpenAI Launches GPT-6.2: What Our Evaluations Revealrogo-news1794944100895469746, 1794944100899469748, 1794944100902469750, 179494410090346975111-17 00:00
1794993392198923756GPT-6.2 Sol and Astra API: Free Trials and Discounted Accessfelo-blog179499366505892617411-18 00:00

Related stories