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

🔍 Read the full analysis: GPT-6.2 Astra And Sol: What’s Changed In Cost, Scores, And Work? on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, alongside a legal configuration called Astra for Law. Artificial Analysis placed Astra near the top of its Intelligence Index and found lower per-task costs than several rivals, while reporting no improvement in its Coding Agent Index. OpenAI rated both models Critical for cybersecurity.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, with a third offering, Astra for Law 6.2, listed as a legal configuration. Same-day evaluations by Artificial Analysis put Astra among the strongest model families it tested and reported lower cost per task than several competitors, while OpenAI’s safety report assigned both new models a Critical cybersecurity rating.

On the Artificial Analysis Intelligence Index, Astra scored 62 at its Max setting and Sol scored 57. The index leader, Claude Fable 5.2, scored 65 in the family comparison. Astra’s unrounded score was 62.14, just below Fable 5.2 at 62.22 in the comparison of those configurations; Astra’s model page ranks it third of 91 configurations. Artificial Analysis said Astra gained five points over GPT-6 Astra, while Sol gained two points and scored about the same as the older model.

Artificial Analysis estimated that a task in its index cost $4.00 with Astra Max and $1.01 with Sol Max. The corresponding estimates were $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 general prices for every workload. OpenAI lists Astra at $8 per million input tokens and $40 per million output tokens, down from GPT-6 Astra’s $10 and $50. Sol is priced at $2 per million input tokens and $10 per million output tokens.

The models share a 1,050,000-token context window, can accept up to 922,000 input tokens and can return up to 128,000 output tokens, according to OpenAI’s developer materials. Astra is positioned for the most demanding work; Sol is presented as a lower-cost option for complex tasks and agents. The pricing page lists Astra for Law at the same standard rates as Astra, but the available documentation does not detail how the legal configuration differs.

At a glance
reportWhen: Released and evaluated November 17, 2026
The developmentOpenAI released GPT-6.2 Astra and Sol, with same-day independent evaluations and a new safety report showing strong scores alongside cost and risk trade-offs.

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

Costs and Capabilities Vary by Workload

The release gives developers a choice between a higher-scoring model and a lower-priced one, but the index figures do not settle which is cheaper or better for a particular application. Artificial Analysis’s per-task estimates reflect its own benchmark workload; actual bills depend on prompt size, output length, processing speed and request type. That distinction matters for teams comparing a model’s list price with the cost of completing a real job.

The results are also not a across-the-board advance. Artificial Analysis reported no improvement in its Coding Agent Index, and measured hallucination rates of 59% for Astra and 55% for Sol at maximum effort. Those results are specific to the cited evaluations and settings, but they caution against treating a strong overall index score as proof of consistent performance in coding or factual tasks.

OpenAI’s safety report adds a separate deployment consideration: both models meet the company’s Critical cybersecurity threshold, its higher of two risk thresholds. OpenAI says more permissive cyber access requires extra vetting. Organizations considering the models for software development, security-related work or agent systems will need to weigh the reported capabilities against those safeguards and their own testing requirements.

KitchenAid Artisan Mini Tilt-Head Stand Mixer, 3.5 Quart, Aqua Sky Blue

KitchenAid Artisan Mini Tilt-Head Stand Mixer, 3.5 Quart, Aqua Sky Blue

De'Longhi Stilosa Compact Manual Espresso Machine with Manual Milk Frother

De'Longhi Stilosa Compact Manual Espresso Machine with Manual Milk Frother

As an affiliate, we earn on qualifying purchases.

How OpenAI Positions the Models

OpenAI describes Astra as its model for demanding projects and Sol as its option for everyday work, including coding and agents at scale. Both accept text and images and produce text, according to the developer guide. Their knowledge cutoffs differ: May 18, 2026, for Astra and June 1, 2026, for Sol. OpenAI says fine-tuning is not supported.

Listed rates are not the only pricing factor. For requests with more than 272,000 input tokens, OpenAI says input and cached-input rates double and output rates rise to 1.5 times the standard short-context rate for the full request. Batch processing costs half the standard rate, while Fast costs twice as much. Ultrafast is listed at three times standard pricing through February 28, 2027, and six times from March 1. Eligible background requests receive lower rates depending on their completion window.

OpenAI also documents an API behavior change: from GPT-6.2, editing conversation history mid-conversation drops the model’s earlier reasoning, which may lower answer quality and lead to repeated work and more tokens. The API rejects requests that set reasoning effort to “none.” The source report says its review of available pages could not establish which ChatGPT plans include the models.

Open Questions on Access and Safety

Several details remain unresolved in the source materials. The reviewed OpenAI pages did not establish which ChatGPT plans include Astra or Sol, and they do not explain the technical differences between Astra and Astra for Law. Artificial Analysis had not published an Intelligence Index score for the legal configuration, although it had tested it on one legal benchmark.

The independent scores and cost estimates are tied to named tests and model settings; they do not predict results across every task, and the source report provides no complete benchmark breakdown in the material available here. OpenAI’s Critical cybersecurity designation is a risk rating, not evidence that either model has caused a security incident. The practical effect of its vetting requirements, and how model performance compares across real deployments, remains to be seen.

Watch for Access and Evaluation Details

Developers can compare the models through OpenAI’s API, subject to account-tier rate limits and the listed processing and data-residency terms. OpenAI says approved organizations can use Zero Data Retention and US or EU data residency with Standard, Fast and Batch processing; Ultrafast is limited to US data residency. Promotional pricing for Astra for Law is listed as available at least through December 31, 2026.

Further clarity will depend on OpenAI’s product and documentation updates, including ChatGPT availability, the distinction between Astra and Astra for Law, and any additional safeguards for cyber-related use. Independent evaluations may also add results for the legal configuration and broader evidence on coding, hallucination rates and real-world task costs.

Key Questions

How did GPT-6.2 Astra and Sol score?

Artificial Analysis reported Intelligence Index scores of 62 for Astra Max and 57 for Sol Max. Claude Fable 5.2 led the family comparison at 65. Astra’s Max configuration ranked third on its model page, with an unrounded score just below Fable 5.2’s tested configuration.

Which model costs less?

Sol has lower listed API rates: $2 per million input tokens and $10 per million output tokens, compared with Astra’s $8 and $40. Artificial Analysis also estimated a lower cost for its benchmark task with Sol Max, but actual costs vary with workload and API settings.

Did coding performance improve?

Not in the specific measure cited. Artificial Analysis reported no improvement in its Coding Agent Index. That finding applies to that evaluation and does not establish how the models perform on every coding task.

What does OpenAI’s Critical cybersecurity rating mean?

OpenAI’s system card says both models meet its Critical threshold for cybersecurity, the higher of the two risk thresholds described in the source. OpenAI says more permissive cyber access requires extra vetting; the rating does not by itself indicate that a security incident occurred.

What is GPT-6.2 Astra for Law?

OpenAI lists Astra for Law 6.2 as a legal model configuration with the same standard API rates as Astra. The reviewed documentation does not explain how it differs, and Artificial Analysis had not assigned it an Intelligence Index score.

Source: ThorstenMeyerAI.com

You May Also Like

Claude Sonnet 5.5 Delivers Similar Benchmark Results For Up To 30% Less Per Task

A report says Claude Sonnet 5.5 nearly matches Opus 5.5 on benchmarks at up to 30% lower cost per task, but provides no scores or pricing method.

NVIDIA Nemotron And ByteDance Doubao Come To PATEO’s AI Box

A Yahoo Finance headline links PATEO’s in-vehicle AI Box with NVIDIA DRIVE AGX Thor, Nemotron and ByteDance Doubao; product details remain limited.

iOS 27, iPadOS 27, And macOS 27

Apple is expected to announce iOS 27, iPadOS 27, and macOS 27, with search interest spiking amid rising speculation about new features and hardware integration.

How Ring’s Flying Home Drone And Patrol Robot May Fit Into Everyday Life

Ring announced an indoor drone and a business patrol robot, with different launch dates, prices and privacy questions.