What’s New For Work With GPT-6.2 Astra And Sol?
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: What’s New For Work With GPT-6.2 Astra And Sol? 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 for demanding tasks and lower-cost GPT-6.2 Sol for everyday work on November 17, 2026, alongside a legal configuration. Independent tests place Astra near the top of a leading model index, while reporting no improvement in its Coding Agent Index; OpenAI rates both models Critical for cybersecurity.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, presenting Astra for demanding projects and Sol as a lower-cost option for everyday work and agents. The launch also includes Astra for Law 6.2. Same-day independent testing puts Astra among the leading model families, but reports no improvement on a coding-agent index; OpenAI’s safety report rates both new models Critical for cybersecurity.

OpenAI describes Astra as its model for its hardest projects and Sol as its go-to for everyday work, including coding and powering agents at scale. Its developer guide recommends Astra for the most demanding tasks and Sol for complex work at lower cost. Both models process text and images, have a 1,050,000-token context window, accept up to 922,000 input tokens and can return up to 128,000 output tokens. Fine-tuning is not supported. Astra’s knowledge cutoff is May 18, 2026; Sol’s is June 1, 2026.

On Artificial Analysis’s Intelligence Index, accessed November 17, Astra scores 62 at its Max setting and Sol scores 57. Claude Fable 5.2 leads the model-family comparison with 65. The scores are close: Astra’s unrounded score is 62.14, compared with 62.22 for Fable 5.2 at Xhigh, placing Astra third among individual configurations on its model page. Artificial Analysis says Astra improved five points over GPT-6 Astra and Sol improved two; it also reports no improvement in the Coding Agent Index.

The same index’s task-cost comparison favors the new OpenAI models on the specified task: $4.00 for Astra Max and $1.01 for Sol Max, against $26.54 for Claude Fable 5.2 Max and $5.48 for Gemini 4 Carbon High. Those figures apply to that benchmark task, not every workload. OpenAI lists Astra at $8 per million input tokens and $40 per million output tokens, 20% below the previous GPT-6 Astra prices; Sol costs $2 and $10. Cached input is priced at $0.40 per million tokens for Astra and $0.10 for Sol.

OpenAI also lists GPT-6.2 Astra for Law as a legal model configuration at the same standard rates as Astra. Its model ID is gpt-6.2-astra-law. Artificial Analysis has not published an Intelligence Index score for it, though it has tested the model on one legal benchmark. OpenAI’s API documentation says that, from GPT-6.2 onward, editing conversation history midstream removes the model’s earlier reasoning; answer quality may fall and the model may use more tokens repeating work. API requests setting reasoning effort to “none” are rejected.

At a glance
announcementWhen: Announced November 17, 2026; independen…
The developmentOpenAI released GPT-6.2 Astra, GPT-6.2 Sol and a legal configuration, with same-day independent tests and a safety report showing strong but mixed results.

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

The Cost and Safety Trade-Off

The release gives businesses and developers a new choice between a higher-performing model and a cheaper one, with independent results suggesting the price difference can matter at scale. Artificial Analysis reported that Sol reached a score on par with GPT-6 Astra at about 15% of its cost per task in the index comparison. That does not establish that Sol is the better choice for every application: the index covers a defined set of tests, and performance can differ across real workloads.

Cost also depends on how an API request is handled. For prompts above 272,000 input tokens, OpenAI doubles input and cached-input rates and charges output at 1.5 times the short-context rate for the entire request. Batch processing costs half the standard rate; Fast costs twice as much. Ultrafast costs three times the standard rate through February 28, 2027, then six times from March 1. Eligible background requests receive lower token rates, subject to two-hour or eight-hour completion windows. Teams comparing options need to account for these conditions as well as headline prices.

Safety findings matter for organizations considering agentic or sensitive work. OpenAI’s system card rates both models Critical for cybersecurity, its higher of two cyber-risk thresholds. It says more permissive cyber access requires additional vetting. The report also identifies GPT-6.2 Astra as the first model OpenAI has released externally that meets its High threshold for AI self-improvement. That is OpenAI’s assessment under its own framework, not a finding that the model has caused harm.

Breville BES870BSXL Barista Express Espresso Machine, Black Sesame

Breville BES870BSXL Barista Express Espresso Machine, Black Sesame

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Bose QuietComfort Wireless Noise-Cancelling Headphones, Sandstone

Bose QuietComfort Wireless Noise-Cancelling Headphones, Sandstone

As an affiliate, we earn on qualifying purchases.

Pricing, Limits and Access

OpenAI’s standard API rates make Astra less expensive than its predecessor, GPT-6 Astra, which is listed at $10 per million input tokens, $1 for cached input and $50 per million output tokens. The new Astra is listed at $8, $0.40 and $40, respectively; Sol is listed at $2, $0.10 and $10. Astra for Law has the same listed rates as Astra. OpenAI says promotional pricing for the legal configuration is available at least through December 31, 2026.

Listed API rate limits vary by usage tier, from 5,000 requests and 1 million tokens per minute to 15,000 requests and 40 million tokens per minute. Approved organizations can use Zero Data Retention and US or EU data residency with Standard, Fast and Batch processing; Ultrafast is listed with US data residency only. OpenAI’s launch post was not accessible to the source author, so the information available does not establish which ChatGPT plans include the models. Pricing and API availability should not be taken as confirmation of consumer-plan access.

Artificial Analysis evaluates models across nine tests and reports scores, task costs and completion times. Its same-day assessment says OpenAI now trails only Anthropic’s Claude Fable 5.2 in the headline model-family comparison. The result is not a sweep: Astra leads some tests and trails on others, while the reported coding-agent result shows no improvement. The source article was researched and drafted with AI assistance, including Anthropic’s Claude, which competes with OpenAI; it says neither OpenAI, Anthropic nor Artificial Analysis reviewed or sponsored the article.

Questions on Access and Performance

The available information does not establish which ChatGPT plans include Astra or Sol, because the launch post could not be accessed. OpenAI’s listing of an API model and prices does not answer that consumer-access question. It is also unclear from the available pages how Astra for Law differs from standard Astra beyond its legal configuration label and model ID.

Artificial Analysis has not assigned Astra for Law an Intelligence Index score, and the source material describes only one legal benchmark test for it. Broader performance in legal work is therefore not established by the cited results. More generally, index scores and the reported cost per task describe specific evaluation conditions; they do not determine outcomes or costs for every customer workload.

OpenAI’s cyber classifications and self-improvement assessment come from its own system card. The source material does not provide enough detail to independently establish how those thresholds compare with other companies’ safety frameworks. The index also reports hallucination rates of 59% for Astra and 55% for Sol at maximum effort, but the source excerpt does not give further detail here on the test setup or a comparison baseline. Those figures should be read as reported test results, not as a universal rate for everyday use.

What Developers Should Check

Developers considering the models can compare them on their own tasks, including output quality, token use, latency and the API processing tier needed. They should account for the higher rates on prompts exceeding 272,000 input tokens and the documented behavior change when conversation history is edited. OpenAI’s published tier limits and data-residency options provide additional details for deployment planning, subject to account eligibility.

Further independent evaluations could clarify Astra’s performance across coding, science and legal work, while OpenAI’s model pages or launch materials may answer questions about ChatGPT plan access and the legal configuration. The legal configuration’s promotional pricing is listed through at least December 31, 2026. Ultrafast pricing is scheduled to rise on March 1, 2027, after a temporary three-times-standard rate through February 28. Until additional details or tests appear, the confirmed picture is a high-ranking release with lower listed prices than its predecessor, mixed benchmark results and significant cyber-risk ratings.

Key Questions

What did OpenAI release?

OpenAI released GPT-6.2 Astra for demanding projects and GPT-6.2 Sol for everyday work and lower-cost complex tasks on November 17, 2026. It also listed GPT-6.2 Astra for Law as a legal configuration.

Which model scored higher in the reported index?

Astra scored 62 at Max and Sol scored 57 on Artificial Analysis’s Intelligence Index. Claude Fable 5.2 led the model-family comparison at 65; Astra’s unrounded score was narrowly below Fable 5.2’s Xhigh configuration.

Are the new models cheaper than GPT-6 Astra?

OpenAI lists Astra at $8 per million input tokens and $40 per million output tokens, compared with $10 and $50 for GPT-6 Astra. Sol is listed at $2 and $10. Long prompts and faster processing tiers can change the final API bill.

What did OpenAI’s safety report say?

OpenAI’s system card rates both GPT-6.2 models Critical for cybersecurity and says more permissive cyber access requires additional vetting. It also says Astra meets OpenAI’s High threshold for AI self-improvement.

Are Astra and Sol included in ChatGPT plans?

The source material does not confirm which ChatGPT plans include the models. OpenAI’s API pricing and model listings alone do not establish consumer-plan availability.

Source: ThorstenMeyerAI.com

You May Also Like

2026’S Ultimate Mesh WiFi Systems For Whole-Home Coverage

Explore the latest mesh WiFi systems for 2026, featuring WiFi 6 and WiFi 7 options, covering large homes with speed, reliability, and future-proof features.

How Desktop CNC Machines Open Up Precision Making

Why are desktop CNC machines transforming craftsmanship, and how can they unlock new levels of precision for your projects? Keep reading to find out.

Apple’s Mixed-Reality Glasses Gain Traction, Blurring Lines Between Real and Virtual

Luring users with seamless AR and VR experiences, Apple’s mixed-reality glasses are revolutionizing digital interaction—discover what makes them so compelling.

AI “Hallucinations” Aren’t Random—Why Models Confidently Invent Facts

Because AI models predict text based on patterns, they confidently invent facts, leaving us to wonder how training data influences these hallucinations.