openai.com web signal

OpenAI Ships GPT-6.2 Astra at $8/$40, 20% Below GPT-6 Astra

6 sources tracking this story

TL;DR

  • GPT-6.2 Astra's $8/$40 per million tokens matches Claude Fable 5.2's list rate, shifting competition entirely to token efficiency per task.
  • OpenAI claims one-third the cost per task versus GPT-6 Astra; independent legal benchmarks rank the new model 11th-30th on that vertical.
  • Third-party evaluators recorded 55-59% hallucination rates across both models, a material risk for the law and finance deployments OpenAI is targeting.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, dropping Astra's standard price to $8 per million input tokens and $40 per million output, a 20% cut from GPT-6 Astra, while Sol holds its predecessor's $2 and $10 rates. Cached input drops to $0.40 for Astra and $0.10 for Sol. In its launch post, the company positions Astra for documents, slides and coding, and Sol as the everyday model that 'lets you complete more tasks on the same paid plan.'

Both models carry a 1,050,000-token context window, accept text and images, and shipped on day one inside ChatGPT Work, Codex and the API for Plus, Pro, Enterprise and Edu users. On Cognition's FrontierCode 2.0, Sol finished within half a point of Astra at about a fifth of the cost, and Cognition put the model into Devin at launch.

Artificial Analysis said on X that Astra 'delivers the #2 model in the Artificial Analysis Intelligence Index and most of the Cost per Task Pareto frontier,' pointing to roughly 78% on Agents' Last Exam v1.1 at about $3 per task and around 83% on BenchCAD 2.1 Core at $5.50 per task. Sol trails Astra by a few points on the same evals at a markedly lower per-task cost. The same scorecard clocks Anthropic's Claude Fable 5.2 at around $60 per task on Agents' Last Exam.

OpenAI's post does not detail training compute or how Ultrafast decode is rate-limited across tiers.

What others are reporting

Coverage cluster as of 2h after publish

  1. The Decoder Read →

    Independent evals show legal rankings of 11th-30th and hallucination rates of 55-59%, undercutting the professional-work claim. Flags hidden chain-of-thought as an enterprise compliance risk.

    Astra is designed for demanding specialized work in science, law, and finance, while Sol is geared toward quick, everyday tasks.
  2. OpenAI Developer Community Read →

    First-party OpenAI announcement confirming API access via gpt-6.2-astra and gpt-6.2-sol, Ultrafast tier availability, and direct pricing and benchmark claims.

    It sets new highs on evals across professional work and scientific research at less than a third of the cost per task of GPT-6 Astra.
  3. 9to5Google Read →

    Frames the release as a direct competitive countermove to Claude Fable 5.2's benchmark dominance, with pricing positioned as a strategic response to losing the performance lead one week earlier.

    GPT-6.2 Astra sets a new quality bar for AI in complex professional work.
  4. Implicator.ai Read →

    Granular cost-per-task analysis beyond headline rates; flags a 22.28% obstacle omission regression that undercuts reliability claims at the professional tier.

    GPT-6.2 Astra now matches Claude Fable 5.2's list price, so any per-task gap comes from how many tokens each model spends.
  5. Digital Applied Read →

    Practical buyer evaluation with concrete testing frameworks for choosing between models based on workflow complexity and risk tolerance, grounded in FrontierCode 2.0 data.

    Sol finished within half a point of Astra on FrontierCode 2.0 at about a fifth of the cost.

Shared on Bluesky by 2 AI experts