AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: GPT-6.2 Astra And Sol At Work: Costs, Scores, And Key Changes on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

  • Fast, free delivery for dorm and study essentials
  • Prime Video and Amazon Music included
  • Member-only deals
Try Prime for Young Adults Free trial for eligible 18–24 year olds
As an affiliate, we earn on qualifying purchases.

TL;DR

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Nov. 17, 2026, alongside a legal configuration called Astra for Law. Artificial Analysis scores Astra at 62 and Sol at 57, with lower reported costs per benchmark task than several rivals; results are mixed, and OpenAI rates both new models at its Critical cybersecurity threshold.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Nov. 17, presenting Astra for demanding work and Sol as a lower-cost option for everyday tasks and agents. Same-day measurements from Artificial Analysis put Astra at 62 points and Sol at 57 on its Intelligence Index, while OpenAI’s safety report rates both at its Critical cybersecurity threshold.

Artificial Analysis describes the release as a strong showing, but not a clean sweep. Its Intelligence Index gives Astra a score of 62 at the Max setting and Sol 57. Claude Fable 5.2 leads the model-family ranking with 65. The more precise unrounded scores show how close the top comparison is: Astra scores 62.14, while Fable 5.2 at Xhigh scores 62.22. Astra’s model page ranks it third among 91 configurations.

On the index’s reported task-cost measure, an evaluated task costs $4.00 with Astra Max and $1.01 with Sol Max. Artificial Analysis lists $26.54 for Claude Fable 5.2 Max and $5.48 for Gemini 4 Carbon High. These are costs for the index’s task, not a universal price for every customer’s use. The same evaluator reports Astra improved five points over GPT-6 Astra, while Sol improved two points and reached a score comparable to GPT-6 Astra at about 15% of its task cost.

OpenAI lists Astra at $8 per million input tokens and $40 per million output tokens, below GPT-6 Astra’s previous $10 and $50 rates. Sol is priced at $2 per million input tokens and $10 per million output tokens. Cached input is listed at $0.40 for Astra and $0.10 for Sol. Those are standard short-context API prices; long requests, speed tiers and eligible background processing have different rates.

At a glance
announcementWhen: Released November 17, 2026; independent…
The developmentOpenAI launched GPT-6.2 Astra and Sol, while same-day independent tests and OpenAI’s safety report outlined their performance, pricing and risks.

GPT-6.2 Astra, Sol and Astra for Law

Three models, three jobs

What launched, what it costs and how it scores in independent tests. Astra and Sol at their Max setting; the better value in each row is outlined.

AstraSolFor Law Price in/out$8/$40$2/$10$8/$40 Index score6257– Cost per task$4.00$1.01– Minutes per task10.47.9– Legal test15.0%11.1%20.8% Both rated Critical for cyber risk
  1. Three releases, one dayOn November 17, OpenAI released GPT-6.2 Astra for the hardest projects and GPT-6.2 Sol for everyday work. Astra for Law 6.2 is a legal configuration at Astra’s price.
  2. The independent indexAstra scores 62, second among model families behind Claude Fable 5.2 at 65. Sol scores 57. One index task costs $4.00 and $1.01.
  3. Speed and a legal testAstra needs 10.4 minutes per task, Sol 7.9. On the Harvey LAB-AA legal test, Astra for Law passes 20.8% of tasks.
  4. OpenAI’s own safety ratingBoth models meet OpenAI’s Critical threshold for cybersecurity. Higher-risk cyber help now needs extra vetting.

Source: Artificial Analysis, model pages for GPT-6.2 Astra, GPT-6.2 Sol and Astra for Law 6.2, accessed Nov 17, 2026 (Intelligence Index, cost and time per task, Harvey LAB-AA v1.1); OpenAI, API pricing page and GPT-6.2 System Card, Nov 17, 2026. Prices per million input/output tokens. Redrawn by Thorsten Meyer AI.

Graphic: Powered by Thorsten Meyer AI

Lower Task Costs, Mixed Results

The launch gives developers a new choice between higher capability and lower cost, with the two models aimed at different workloads. Artificial Analysis’s task-cost figures suggest Sol may be attractive when a lower-cost model can handle the work, while Astra’s score places it close to the index leader. Those measurements are useful for comparison, but actual expenses depend on prompt size, output, processing mode and the developer’s workload.

The performance picture is not uniformly positive. Artificial Analysis reports no improvement in its Coding Agent Index and gives hallucination rates of 59% for Astra and 55% for Sol at maximum effort. Those figures describe results under that evaluator’s tests; they do not establish how either model will perform in every deployment. OpenAI’s own cyber assessment also matters to organizations deciding whether to allow models to access sensitive systems: its report assigns both models the Critical risk rating under the company’s framework.

OpenAI says Astra is the first model it has released externally that meets its High threshold for AI self-improvement. The designation and the cyber ratings are OpenAI’s assessments, not independent findings that the models can independently improve themselves or that a cybersecurity incident has occurred. For customers, the combination of benchmark results, pricing and safety controls makes testing on their own tasks more relevant than relying on a single headline score.

Products Worth Considering

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

As an affiliate, we earn on qualifying purchases.

How the New Models Are Positioned

OpenAI’s developer guide recommends Astra for the most demanding tasks and Sol for complex work at lower cost. Both models accept text and images and return text. They have a listed 1,050,000-token context window, with up to 922,000 input tokens and 128,000 output tokens. The stated knowledge cutoffs differ: May 18, 2026, for Astra and June 1, 2026, for Sol. OpenAI says fine-tuning is not supported.

A third listing, GPT-6.2 Astra for Law, appeared in OpenAI’s legal model configurations on launch day, with its own model ID and the same standard prices as Astra. Artificial Analysis had not assigned it an Intelligence Index score, though it had tested the legal configuration on one legal benchmark. The available pages did not explain how it differs from standard Astra, so its listing alone does not establish what legal-specific changes it contains.

OpenAI also describes an API behavior change beginning with GPT-6.2: when conversation history is edited mid-conversation, the API drops the model’s earlier reasoning. The company warns that this can reduce answer quality and cause the model to spend more tokens repeating work. Requests that set reasoning effort to none are rejected. These details may affect applications that revise prior turns or rely on that setting.

Safety, Pricing and Access Details

Several practical points remain unresolved in the available material. OpenAI’s launch post could not be accessed by the source report, so it did not confirm which ChatGPT plans include the models or when consumer access begins. The public pages also did not explain what distinguishes Astra for Law from Astra, and that legal configuration had no Intelligence Index score at the time of the report.

Benchmark scores and task costs are tied to Artificial Analysis’s stated tests and configurations; they are not guarantees for other prompts or workloads. The report lists Astra’s and Sol’s hallucination rates at maximum effort but does not establish how those rates translate to a particular use. OpenAI’s Critical cybersecurity rating is a risk classification in its system card, not evidence of a specific exploit or attack.

API costs can also differ substantially from the standard rates. Requests with more than 272,000 input tokens have doubled input and cache rates and output rates 1.5 times the standard short-context rate for the whole request. Batch processing is listed at half the standard rate, Fast at twice the rate, and Ultrafast at three times the standard rate through Feb. 28, 2027, rising to six times from March 1. Account-tier rate limits and eligibility for data controls also vary.

Developer Testing and Access

Developers evaluating the models can compare them against their own tasks, checking both quality and total API cost rather than assuming benchmark prices will match their usage. They will also need to account for the changed behavior when conversation histories are edited and for the rejection of requests that set reasoning effort to none.

OpenAI’s published pricing says promotional pricing for Astra for Law is available at least through Dec. 31, 2026. Its listed Ultrafast rate is scheduled to rise on March 1, 2027. Access to the models, account-specific limits and any further information about the legal configuration may become clearer through OpenAI’s product and developer documentation. The source material does not provide a confirmed schedule for those updates.

Key Questions

When did OpenAI release GPT-6.2 Astra and Sol?

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Nov. 17, 2026. Astra for Law 6.2 also appeared in the company’s legal model configurations that day.

Which model scored higher in the reported index?

Astra scored 62 at its Max setting on the Artificial Analysis Intelligence Index, compared with 57 for Sol. Claude Fable 5.2 led the model-family comparison with 65; Astra’s unrounded score of 62.14 was slightly below Fable 5.2’s 62.22 at Xhigh.

How much do Astra and Sol cost through the API?

At standard short-context rates, Astra costs $8 per million input tokens and $40 per million output tokens. Sol costs $2 per million input tokens and $10 per million output tokens. Long prompts and processing tiers can change the bill.

What does OpenAI’s Critical cybersecurity rating mean?

OpenAI’s system card assigns both models its Critical threshold for cybersecurity. This is the company’s risk classification; the supplied material does not report a specific attack or vulnerability arising from the release.

What is different about Astra for Law?

OpenAI listed Astra for Law as a legal model configuration, but the pages available to the source report did not explain how it differs from standard Astra. It had no Intelligence Index score at publication, though it had been tested on one legal benchmark.

Source: ThorstenMeyerAI.com

You May Also Like

The SSD Squeeze: Why Storage Joined The Party

Enterprise and consumer SSD prices surge as AI demand and wafer competition tighten supply, reshaping storage markets in 2026.

Trade Secrets And Tech Trends: Apple’s Bold Move Against OpenAI

Apple has filed a lawsuit against OpenAI, alleging former employees stole trade secrets. This marks a significant legal move in AI industry conflicts.

Without Climate Change, U.S. Heat Wave Called ‘Virtually Impossible’

Research shows that climate change is the primary driver of extreme heat waves in the U.S., making such events nearly impossible without global warming.

See the forecast for what could be some of D.C.’s hottest days on record

Weather forecasts indicate potential for some of Washington D.C.’s hottest days ever, raising concerns about health, infrastructure, and climate impacts.