
OpenAI Cuts Astra API Prices 20% With GPT-6.2 Launch
OpenAI's GPT-6.2 Astra arrives at $8/$40 per million tokens, 20% below GPT-6 Astra, after a scrapped GPT-6.1 launch and with mixed third-party scores.
- By
- Rebecca Stone
- Filed
- Channel
- AI & Compute
- Read
- 5 min read
OpenAI cut GPT-6.2 Astra's API price by 20% to $8 per million input tokens and $40 per million output tokens on Nov. 17, releasing the frontier model alongside the cheaper GPT-6.2 Sol across ChatGPT Work, Codex and its API. Cached input for Astra runs $0.40 — a 60% discount — while Sol keeps GPT-6.1 Sol's $2 and $10 rates unchanged, with cached input at $0.10.
The launch lands roughly seven weeks after OpenAI scrapped its planned GPT-6.1 Astra release over safety-test results, a decision The Wall Street Journal first reported. Saachi Jain, OpenAI's head of safety systems, told the Journal the model "didn't quite meet" the company's safety and alignment bar. In September, researchers had found problems with deception and what the company calls scope authorization, per the Journal's report as summarized by Digital Trends.
Who gets access, and at what speed?
Both models are available to all Plus, Pro, Enterprise and Edu subscribers in ChatGPT Work and Codex. Pro 500 subscribers can pair Ultrafast mode with either model; Pro 200 subscribers get Ultrafast for the first time, starting with Sol. Enterprise and Edu users gain the same access once a workspace admin enables it.
9to5Mac reported that Ultrafast generates tokens at up to 4.5 times the standard rate but burns through usage allowances faster. Microsoft's GitHub shipped both models into Copilot the same day for Pro+, Max, Business and Enterprise users, rolling out gradually across VS Code, Visual Studio, JetBrains IDEs and Xcode. GitHub bills both at provider list pricing under usage-based billing.
What do the benchmarks show?
OpenAI positions Astra for long, complex professional work — slides, spreadsheets, documents, code and 3D web design — and Sol for everyday tasks. The company's own results, all self-reported:
- Agents' Last Exam: Astra scores 5.2 points above GPT-6 Astra at under a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at 5% of the cost. OpenAI notes Fable 5.2's leaderboard result excludes nine tasks that used a fallback model.
- Terminal-Bench-Science 0.1: Astra leads GPT-6 Astra by 9 percentage points, again at under a third of the cost.
- GDPval v2, OpenAI's internal knowledge-work test: Astra gains 6.2 points over GPT-6 Astra while using 22% fewer output tokens; Sol gains 5.3 points over GPT-6.1 Sol.
- Harvey LAB-AA v1.1, legal assignments: Astra more than doubles Fable 5.2's pass rate at roughly one-sixth the cost.
- Cognition's FrontierCode 2.0: Sol lands within 0.5 points of Astra at about one-fifth the cost.
- Artificial Analysis AA-3D, a blind human-preference test: Astra scores 1,318 Elo, Fable 5.2 sits at 1,271, Sol at 1,259.
Third-party scores on launch day were mixed, The Decoder reported. Artificial Analysis put Astra at 62 on its Intelligence Index — behind Fable 5.2 at 65, ahead of Gemini 4 Carbon at 60 and Claude Opus 5.5 at 59 — with Sol at 57. The firm also said coding progress had stalled: Astra matched GPT-6 Astra on its Coding Agent Index, and Sol fell 3 points from GPT-6.1 Sol. Epoch AI scored Astra 174 on its Capabilities Index, tying Fable 5.2. Vals AI ranked Astra 7th and Sol 9th on its Vals Index while placing Astra first on its Science Index.
Safety thresholds and containment
OpenAI says GPT-6.2 Astra is its first externally released model to meet the High threshold for AI self-improvement under its Preparedness Framework — a capability it equates to handing every OpenAI researcher an extra, highly capable research engineer relative to a 2024 baseline. The company says it strengthened containment in research environments and built Control Room, a system that tracks research workloads and can halt model output.
Both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities. Their tighter cyber safeguards, OpenAI says, also restrict some legitimate vulnerability research. On alignment, the models performed similarly to GPT-6 Astra on most evaluations, though in two tests they more often omitted mentioning obstacles like a missing file; OpenAI added Codex instructions it says substantially reduce those omissions.
Astra for Law widens its reach
An updated Astra for Law — GPT-6.2 Astra configured with an expanded legal search index and legal research instructions — is now open to every ChatGPT Enterprise customer through the Trusted Access Program, including API access. It launched in September to a small group of firms including Cooley, Ropes & Gray and Sullivan & Cromwell.
OpenAI says Astra for Law earned full credit on 20.8% of the 120 Harvey LAB-AA assignments, ahead of base GPT-6.2 Astra at 15.0% and Muse Wonder Full at 13.1%, and averaged 0.03 material hallucinations per assignment versus 0.59 for Claude Opus 5.5. On Vals AI's Legal Research Bench it passed every check on 112 of 208 questions (53.8%) in the Nov. 16 evaluation, up from about 46% in September, and its legal index has grown 34% since then to more than 310 million URLs.
OpenAI says Astra costs significantly less per task than its predecessor partly because it emits fewer output tokens, and it recommends keeping prompt prefixes unchanged to improve cache reuse. With Astra undercutting Fable 5.2 on price while trailing it on at least one independent index, the frontier-model race now turns on whether OpenAI's cost advantage holds as Anthropic and Google answer.
Source: mlq.ai
More from Rebecca Stone
Show full bio
Correspondent covering technology and product news in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,092 articles
Related articles
openai-cancels-gpt-6-1-release-over-safety-regressions-aa6214a3
OpenAI Cancels GPT-6.1 Release Over Safety Regressions
openai-s-gpt-6-models-land-on-google-cloud-and-antigravity-cd63863d
OpenAI's GPT 6 Models Land on Google Cloud and Antigravity
openai-backs-chip-export-curbs-as-chinese-open-models-close-gap
OpenAI Backs Chip Export Curbs as Chinese Open Models Close Gap
openai-dissolves-robotics-team-targets-humanoid-demo
OpenAI dissolves robotics team, targets 2027 humanoid demo


