OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, pricing gpt-6.2-astra at $8 input and $40 output per million tokens and keeping gpt-6.2-sol at $2 input and $10 output. Both models launched in ChatGPT Work, Codex, and the OpenAI API, with GPT-6.2 Astra built for complex projects and GPT-6.2 Sol built for everyday work.
OpenAI announced the two new models on its official release page, followed by an announcement post from OpenAI on X at 17:54 UTC and a developer post from OpenAIDevs at 18:02 UTC on November 17, 2026. All availability, pricing, and benchmark details below reflect that announcement as checked on November 18, 2026.
What did OpenAI release on November 17, 2026?
GPT-6.2 Astra is OpenAI’s new frontier model for the hardest projects, while GPT-6.2 Sol is OpenAI’s efficient, versatile model for everyday work. Across both releases, OpenAI states that the models deliver stronger results on demanding tasks using fewer output tokens, helping teams accomplish more inside existing usage limits and reach frontier performance at costs suited to production scale.
OpenAI positions GPT-6.2 Astra around complex professional deliverables. According to the release page, OpenAI trained GPT-6.2 Astra to produce clearer slides, spreadsheets, and documents that follow user templates and domain conventions in fields such as law, sales, finance, and accounting, with more direct writing and less filler. OpenAI also states that GPT-6.2 Astra writes cleaner code, builds better-performing frontends for websites, gaming, and 3D design, and represents what OpenAI believes is the best model in the world for scientists.
OpenAI positions GPT-6.2 Sol as its go-to model for daily technical and non-technical work. Compared with GPT-6.1 Sol, GPT-6.2 Sol is described as smarter and more efficient, enabling users to finish more tasks on the same paid plan while handling multistep agentic workflows and natural collaboration. OpenAI notes that many of its own employees use GPT-6.2 Sol as their default model for both technical and non-technical tasks, and that its speed and capability also suit demanding personal tasks outside of work.
OpenAI’s November 17, 2026 announcement introduces only GPT-6.2 Astra and GPT-6.2 Sol. The release page makes no mention of a GPT-6.2 Terra or a GPT-6.2 Luna, and it gives no indication that those earlier tiers were updated alongside this launch. If you are looking for how the earlier generation was structured across flagship, everyday, and lightweight tiers, see our GPT-6 Astra vs GPT-6.1 Sol vs Luna comparison.
Where can you use GPT-6.2 Astra and Sol today?
GPT-6.2 Astra and GPT-6.2 Sol launched on November 17, 2026 exclusively inside ChatGPT Work, Codex, and the OpenAI API for eligible users. On the product side, Plus, Pro, Enterprise, and Edu subscribers gain access to both models in ChatGPT Work and Codex.
Access rules differ between individual paid plans and managed organization workspaces:
- Plus and Pro accounts: Users on Plus and Pro plans can select GPT-6.2 Astra and GPT-6.2 Sol in ChatGPT Work and Codex starting on launch day.
- Enterprise and Edu workspaces: Eligible Enterprise and Edu users receive identical access in ChatGPT Work and Codex once their workspace administrator enables the models.
- Ordinary ChatGPT consumer chat: OpenAI’s announcement scopes launch access to ChatGPT Work and Codex and does not describe GPT-6.2 Astra or GPT-6.2 Sol as the default model for ordinary ChatGPT consumer chat.
- OpenAI API: Developers can invoke
gpt-6.2-astraandgpt-6.2-soldirectly through the API at launch.
OpenAI’s November 17, 2026 announcement does not list Free, Go, or Team plan rollout details for GPT-6.2 Astra or GPT-6.2 Sol as of November 18, 2026.
How much do the GPT-6.2 APIs cost?
OpenAI’s official API pricing as of November 18, 2026 reduces token rates for gpt-6.2-astra relative to GPT-6 Astra while keeping gpt-6.2-sol at GPT-6.1 Sol rates. Specifically, gpt-6.2-astra carries a 20% discount on input and output tokens and a 60% discount on cached input tokens compared with GPT-6 Astra.
| Model (API ID) | Input (per 1M tokens) | Cached Input (per 1M tokens) | Output (per 1M tokens) | Pricing Change vs Previous Tier |
|---|---|---|---|---|
GPT-6.2 Astra (gpt-6.2-astra) |
$8 | $0.40 | $40 | 20% lower input/output and 60% cached input discount vs GPT-6 Astra |
GPT-6.2 Sol (gpt-6.2-sol) |
$2 | $0.10 | $10 | Keeps GPT-6.1 Sol API token rates |
Table: Official OpenAI API list prices from the November 17, 2026 announcement (checked November 18, 2026).
On per-task economics, OpenAI states that GPT-6.2 Astra costs significantly less per task than GPT-6 Astra because lower token rates combine with improved token efficiency. To help engineering teams take advantage of the $0.40 cached input rate on gpt-6.2-astra and the $0.10 cached input rate on gpt-6.2-sol, OpenAI recommends keeping prompt prefixes unchanged and appending new instructions to the conversation, which improves cache reuse, reduces cost and latency, and preserves reasoning across turns.
OpenAI did not publish separate batch, audio, or image pricing rates on the November 17, 2026 announcement page. For broader context across providers and token tiers, you can check our AI API pricing comparison.
What does Ultrafast cover on Pro 500, Pro 200, and the API?
Ultrafast availability for GPT-6.2 Astra and GPT-6.2 Sol depends on whether you use ChatGPT Pro 500, ChatGPT Pro 200, or the OpenAI API. OpenAI confirmed three launch rules in its pricing and availability section as of November 18, 2026:
- Pro 500 subscribers can use Ultrafast with either GPT-6.2 Astra or GPT-6.2 Sol.
- Pro 200 subscribers gain Ultrafast access for the first time, starting with GPT-6.2 Sol.
- API developers have Ultrafast available for both
gpt-6.2-astraandgpt-6.2-solat launch.
OpenAI’s November 17, 2026 announcement page did not publish speed multipliers or credit consumption multipliers for Ultrafast mode on either plan or in the API. Any numerical speed or credit multiplier figures circulating elsewhere were not part of the official release page when checked on November 18, 2026.
How do official evals compare Astra, Sol, and Claude Fable 5.2?
OpenAI’s November 17, 2026 announcement reports internal and third-party evaluation results comparing GPT-6.2 Astra and GPT-6.2 Sol against GPT-6 Astra, GPT-6.1 Sol, and Anthropic’s Claude Fable 5.2. Lotook has not independently tested either GPT-6.2 model, so every figure below reflects OpenAI’s published numbers and footnotes as of November 18, 2026.
Long-horizon knowledge work, legal briefs, and data analysis
GPT-6.2 Astra shows its largest reported margins over prior models and Claude Fable 5.2 on multi-step professional and legal benchmarks:
- Agents’ Last Exam: OpenAI reports that GPT-6.2 Astra sets a new high, scoring 5.2 points higher than GPT-6 Astra at less than one-third the cost, and 5.5 points ahead of Claude Fable 5.2 at 5% of the cost. OpenAI includes an explicit caveat that the Fable 5.2 leaderboard results exclude nine tasks that used a fallback model.
- Harvey LAB-AA v1.1: On legal assignments requiring models to extract facts across source documents and produce error-free deliverables, GPT-6.2 Astra more than doubles Claude Fable 5.2’s pass rate at roughly one-sixth the cost, with OpenAI’s Astra for Law configuration scoring higher still. OpenAI notes that the headline score awards credit only when a deliverable meets every grading criterion with zero material hallucinations.
- GDPval v2: On OpenAI’s internal evaluation of long-horizon workflows across sectors including consulting, finance, design, data science, biotechnology, and marketing, GPT-6.2 Astra leads tested models, gaining 6.2 percentage points over GPT-6 Astra while using 22% fewer output tokens. GPT-6.2 Sol improves 5.3 points over GPT-6.1 Sol.
- Artificial Analysis AA-AnalystAgent: On spreadsheet and report analysis tasks such as calculating medical transport spending from California Medicaid expenditure reports, GPT-6.2 Astra sets a new high by answering 62.5% of questions correctly on all five independent attempts.
Coding, science, 3D design, and everyday tasks
OpenAI’s published evaluations highlight a clear split between code-merging efficiency on Sol and scientific or 3D specialization on Astra:
- Cognition FrontierCode 2.0: Evaluating whether changes to real codebases meet maintainers’ merge standards, GPT-6.2 Sol comes within 0.5 points of GPT-6.2 Astra at about one-fifth the cost when both models run at max settings.
- Terminal-Bench-Science 0.1: For scientific workflows inside a terminal, GPT-6.2 Astra sets a new high and scores 9 percentage points above GPT-6 Astra at less than one-third the cost. OpenAI footnotes that Fable 5.2 results on this benchmark come from Artificial Analysis in the mini-swe-agent harness, whereas Anthropic reported a 74.6% score for Fable 5.2 at max reasoning in Claude Code, meaning those runs use different harnesses and configurations.
- Vals Science Index: OpenAI states that GPT-6.2 Astra sets a new high on interpreting real biological data and running simulated experiments, though no numeric score was published on the announcement page.
- Artificial Analysis AA-3D: In blinded human comparisons of 3D experiences built from identical prompts, GPT-6.2 Astra ranks first at 1,318 Elo, followed by Claude Fable 5.2 at 1,271 Elo and GPT-6.2 Sol at 1,259 Elo.
- LifeVal: On OpenAI’s benchmark for complex everyday tasks (such as shortlisting kindergartens after a move, matching a home to a family’s budget and commute, or planning meals after a new diagnosis), GPT-6.2 Sol outscores GPT-6 Astra at roughly one-seventh the cost.
OpenAI’s announcement page footnotes the Artificial Analysis Intelligence Index v5 methodology, though it does not publish total Intelligence Index scores for either model. Readers comparing frontier options across labs can also review our coverage of Claude Fable 5.2 and Gemini 4 Carbon.
Early tester reports and safety disclosures
Early partner feedback published by OpenAI points to large token reductions on specialized workflows alongside qualitative gains on slide template matching (preserving layouts, colors, typography down to font size, and writing style from files or screenshots) and complex codebase maintenance. At Basis, Eros Wang reported that average time for long-horizon accounting work dropped 50%, with the most complex work falling up to 73%. At Harvey, Niko Grupen reported more than 3x as many Legal Agent Benchmark tasks completed compared with GPT-6 Astra using 34.5% fewer output tokens. At Legora, Jacob Lauritzen reported 62% fewer tokens than GPT-6.1 Sol on complex legal-agent tasks and stronger results on 8 of 10 legal research tasks across tax, M&A, and litigation. ClickUp’s Jay Hack noted about 40% fewer tokens and 30% fewer review rounds on Workspace builds, while Higgsfield AI’s Alex Mashrabov reported 73% one-shot success in internal 3D generation evaluations versus three or more iterations on GPT-6 Astra.
On safety and alignment, OpenAI states that both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities under its Preparedness Framework, and that GPT-6.2 Astra is OpenAI’s first externally released model to meet the High threshold for AI Self-improvement. Tighter cybersecurity safeguards at launch reduce misuse risk while also restricting some legitimate vulnerability research and security testing workflows, and OpenAI plans to bring GPT-6.2 models to its Daybreak defender tiers. In alignment evaluations, both models took fewer improper actions in simulated workplace tasks than GPT-6 Astra, though in two tests they were more likely to omit mentioning blocking obstacles such as a missing file or unavailable search tool; OpenAI added instructions in Codex that substantially reduce those omissions.
GPT-6.2 Astra vs Sol: which should you pick?
Choosing between GPT-6.2 Astra and GPT-6.2 Sol comes down to workload difficulty, deliverable stakes, and token budget rather than treating one model as strictly better for every job. Because gpt-6.2-sol costs one-quarter as much per token as gpt-6.2-astra and supports Ultrafast on ChatGPT Pro 200, Sol serves as the practical starting point for most teams.
| Priority or Use Case | Recommended Model | Official Basis (as of Nov 18, 2026) |
|---|---|---|
| Everyday work, multistep agentic pipelines, high-volume API calls | GPT-6.2 Sol (gpt-6.2-sol) |
$2 / $10 per 1M tokens; OpenAI’s stated everyday default; outscores GPT-6 Astra on LifeVal at ~1/7 cost |
| Production coding and codebase merges on a budget | GPT-6.2 Sol (gpt-6.2-sol) |
Within 0.5 points of GPT-6.2 Astra on FrontierCode 2.0 at ~1/5 cost (both at max) |
| Pro 200 subscribers needing Ultrafast | GPT-6.2 Sol (gpt-6.2-sol) |
First model to bring Ultrafast to Pro 200; Astra Ultrafast requires Pro 500 or the API |
| Client-ready slides, spreadsheets, and legal or finance deliverables | GPT-6.2 Astra (gpt-6.2-astra) |
Template fidelity, Harvey LAB-AA v1.1 lead, 62.5% 5/5 consistency on AA-AnalystAgent |
| Scientific terminal work, 3D experiences, and complex frontends | GPT-6.2 Astra (gpt-6.2-astra) |
Terminal-Bench-Science 0.1 and Vals Science Index highs; 1,318 Elo on AA-3D |
Start new workflows on gpt-6.2-sol when volume, latency, or per-token cost drives the project, then escalate specific tasks to gpt-6.2-astra where higher template adherence, legal precision, scientific accuracy, or 3D frontend polish justifies the $8 / $40 per-million-token list price. For a broader decision framework across providers and model tiers, see our guide on how to choose an AI model.
FAQ
Are GPT-6.2 Astra and GPT-6.2 Sol the default in standard ChatGPT?
OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, alongside the OpenAI API. As of November 18, 2026, the official announcement does not describe either GPT-6.2 model as the default model for ordinary ChatGPT consumer chat.
Did OpenAI release GPT-6.2 Terra or GPT-6.2 Luna?
OpenAI’s November 17, 2026 announcement introduces only GPT-6.2 Astra and GPT-6.2 Sol. The release page does not mention a GPT-6.2 Terra or a GPT-6.2 Luna, nor does it state that earlier Terra or Luna tiers were updated. Readers tracking the earlier GPT-6 Astra, GPT-6.1 Sol, and Luna lineup can consult our previous family guide.
How do Enterprise and Edu teams enable GPT-6.2 Astra and Sol?
Eligible ChatGPT Enterprise and Edu users get access to GPT-6.2 Astra and GPT-6.2 Sol inside ChatGPT Work and Codex once their workspace administrator enables the models. Unlike Plus and Pro accounts, Enterprise and Edu workspaces do not have the new GPT-6.2 models turned on automatically at launch without admin action.
Does GPT-6.2 Sol include Ultrafast on ChatGPT Pro 200?
GPT-6.2 Sol is the first model to bring Ultrafast to ChatGPT Pro 200 subscribers, as announced on November 17, 2026. ChatGPT Pro 500 subscribers can use Ultrafast with either GPT-6.2 Astra or GPT-6.2 Sol, and API developers can use Ultrafast with both models at launch. OpenAI did not publish speed or credit multipliers.
How can developers lower API costs on gpt-6.2-astra?
Developers calling gpt-6.2-astra pay $0.40 per million cached input tokens instead of the $8 standard input rate, a 60% cached-input discount compared with GPT-6 Astra. To maximize cache reuse, lower latency, and preserve reasoning across turns, OpenAI recommends keeping prompt prefixes unchanged and appending new instructions to the conversation.



