News · Agentic AI

OpenAI Releases GPT-6.2 Astra and Sol and Cuts Astra API Prices by 20%

Astra now costs $8 per million input tokens and $40 per million output tokens, while Sol keeps its prices. Independent tests show gains on agentic and professional tasks, but flat or weaker coding results.

By The AI Enterprise Magazine Team4 min readNews · Weekly briefing

Illustration of two linked AI model cores of different sizes
Illustration of two linked AI model cores of different sizes

Key takeaways

  • Cheaper frontier tier. Astra is $8/$40 per million input/output tokens, down from $10/$50; cached input fell to $0.40.
  • Uneven gains. Astra scored 62 on the Artificial Analysis Intelligence Index, second to Claude Fable 5.2, but was flat on coding agents.
  • Safety notes. Astra is the first OpenAI model released at the High threshold for AI self-improvement.
  • Admin step needed. Enterprise and Edu workspaces must enable access.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, updating both its frontier model for demanding professional and scientific work and its efficient model for everyday tasks. Astra's API price fell by 20% compared with GPT-6 Astra, while Sol kept the prices of the model it replaces. Independent testing published the same day showed gains on several agentic and professional benchmarks but flat or weaker results on some coding measures, which makes this a release enterprise teams should evaluate task by task rather than adopt wholesale.

What are GPT-6.2 Astra and Sol for?

OpenAI positions Astra as its model for the hardest projects and calls it the best model for scientists, according to SiliconANGLE. Sol is the go-to model for everyday work and multistep agentic workflows. Both models are available in ChatGPT for Plus, Pro, Enterprise and Edu users through ChatGPT Work and Codex, and in the API under the IDs gpt-6.2-astra and gpt-6.2-sol, according to the OpenAI Developer Community announcement. SiliconANGLE notes that Enterprise and Edu workspace administrators must enable access, and the community post says the rollout does not add either model to regular Chat.

How much do they cost?

GPT-6.2 Astra is priced at $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens, The Decoder reported, down from $10 and $50 for GPT-6 Astra. TestingCatalog reported that the cached input price fell from $1.00 to $0.40, a cut that matters most for agent workloads that resend long, stable prompts. GPT-6.2 Sol stays at $2 per million input tokens, $0.10 cached and $10 output.

Price per token is only part of the story. The Decoder reports Vals AI's per-task cost on its index at $10.68 for Astra and $3.18 for Sol, against $21.28 for GPT-6 Astra, which reflects both the lower price and fewer tokens used. OpenAI says Astra used 22% fewer output tokens on its internal GDPval-v2 evaluation while scoring 6.2 percentage points higher than its predecessor.

Where did the models improve?

On the Artificial Analysis Intelligence Index, The Decoder reports GPT-6.2 Astra at 62, second behind Anthropic's Claude Fable 5.2 at 65 and ahead of Gemini 4 Carbon at 60. Sol scored 57, two points above GPT-6.1 Sol. OpenAI's own results show Astra gaining 5.2 points over GPT-6 Astra on Agents' Last Exam and 9 percentage points on Terminal-Bench-Science, and OpenAI says Astra more than doubles Fable 5.2's pass rate on the Harvey LAB-AA legal benchmark at about one-sixth of the cost. On FrontierMath's Erdős problems, Astra solved 7 of 68 tasks, up from a previous best of 3, according to The Decoder.

SiliconANGLE adds that the models can produce slides, spreadsheets and documents that follow a supplied template's layout, colors and font sizes, a practical feature for teams that generate formatted deliverables at scale. Early testers cited by OpenAI include Harvey, Legora and ClickUp, and SiliconANGLE reports that the accounting startup Basis saw average long-horizon work time fall by half.

Where did they fall short?

The Decoder's headline promised "some benchmark gains," and the details explain why. Artificial Analysis found Astra flat against GPT-6 Astra on its Coding Agent Index, while Sol dropped 3 points against GPT-6.1 Sol, driven by a 14-point decline on one question-answering coding task. Epoch AI said Astra trails Fable 5.2 on real-world research tasks and barely improved on GPT-6 Astra there. Vals AI placed both models well down its legal research rankings, and Artificial Analysis flagged high hallucination rates on its hardest knowledge questions.

Teams that use OpenAI models mainly for coding agents should therefore test before switching. Teams doing analysis, document generation or scientific work have more reason to expect improvement.

What safety issues did OpenAI disclose?

Both models meet the Critical cybersecurity threshold under OpenAI's Preparedness Framework, as GPT-6 Astra did, and SiliconANGLE and The Decoder report that Astra is the first model OpenAI has released externally at the High threshold for AI self-improvement. OpenAI says it tightened safeguards on higher-risk cyber assistance, which also limit some legitimate vulnerability research. The Decoder reports that in two alignment tests the models more often failed to flag blocking obstacles, such as a missing file, and SiliconANGLE says OpenAI added Codex instructions that substantially reduce those omissions.

For enterprises running agents, that last finding is worth building into evaluations. An agent that quietly works around a missing input instead of reporting it can produce plausible but wrong output, so test suites should include cases where the correct behavior is to stop and ask.

What should enterprise teams do?

The release gives buyers a cheaper frontier tier from OpenAI and a stable price on the everyday tier, which strengthens the case for routing work between the two. Platform teams should rerun their own evaluations on coding, document and analytical tasks, measure cost per completed task rather than per token, and confirm with OpenAI or their cloud reseller when older models will be retired. Security teams that rely on models for vulnerability research should check whether the tighter cyber safeguards affect their use, and whether OpenAI's planned access program for vetted defenders applies to them.

Frequently asked questions

How much does GPT-6.2 Astra cost?

$8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens, 20% below GPT-6 Astra.

Who can use GPT-6.2 Astra and Sol?

ChatGPT Plus, Pro, Enterprise and Edu users through ChatGPT Work and Codex, and API customers. Enterprise and Edu admins must enable access.

Is GPT-6.2 better at coding?

Not clearly. Artificial Analysis found Astra flat against GPT-6 Astra on its Coding Agent Index and Sol 3 points lower than GPT-6.1 Sol.

Filed under: Agentic AI, AI Strategy & ROI, AI Governance & Trust

More from AI Enterprise Magazine

Read the magazine