OpenAI releases GPT-6.2 Astra and Sol, prices Astra 20% below predecessor
OpenAI says both models beat their predecessors at lower cost. Independent tests rank Astra near the top but flag flat coding scores and a high rate of guessing.
Key takeaways
- OpenAI released GPT-6.2 Astra and GPT-6.2 Sol in ChatGPT Work, Codex and the API on Nov. 17.
- OpenAI says Astra's API price is 20% below its predecessor's, at $8 per million input and $40 per million output tokens; Sol's price is unchanged.
- Artificial Analysis ranks Astra second on its intelligence index, behind Claude Fable 5.2, but finds coding scores flat.
- OpenAI says both models hit its Critical cybersecurity threshold and is adding stricter safeguards.
- OpenAI says Astra is the first of its public models to cross its High threshold for AI self-improvement.
OpenAI launched a pair of models, GPT-6.2 Astra and GPT-6.2 Sol, on Tuesday, Nov. 17, according to The Decoder and GitHub. Astra is aimed at demanding professional and scientific work, while Sol is meant as a faster, cheaper model for everyday tasks, The Decoder reports.
Both are available to ChatGPT Work and Codex users on the Plus, Pro, Enterprise and Edu plans, and to developers through the OpenAI API, The Decoder says. GitHub said both are now generally available in GitHub Copilot for Pro+, Max, Business and Enterprise users, with a gradual rollout.
What do they cost?
OpenAI puts Astra's API price at $8 per million input tokens and $40 per million output tokens, 20% below its predecessor, according to The Decoder. Sol keeps the price of GPT-6.1 Sol, at $2 per million input tokens and $10 per million output tokens.
What does OpenAI claim?
OpenAI says both models do better than their predecessors on benchmarks while costing less to run, The Decoder reports. According to OpenAI, Astra handles professional conventions in law, finance and accounting more reliably, and is better at computer use. These are the company's own results.
GitHub, which offers the models in Copilot, said that in its early testing both found clearly more genuine problems during code review than previous GPT-6 models, and used fewer tokens doing it.
What do independent testers say?
Outside evaluators give a mixed picture, according to The Decoder. Artificial Analysis and Vals AI both found Astra leading on science benchmarks. On the Artificial Analysis Intelligence Index, Astra scores 62 and ranks second, behind Anthropic's Claude Fable 5.2 at 65, while Artificial Analysis says OpenAI models lead on cost versus performance. The same group found coding results flat: Astra's coding agent score stayed level with GPT-6 Astra's, and Sol fell three points from GPT-6.1 Sol. Artificial Analysis also measured a high rate of guessing on very hard factual questions, 59% for Astra and 55% for Sol at maximum effort, though The Decoder notes those questions do not reflect everyday use.
Epoch AI found Astra tied with Claude Fable 5.2 on its capabilities index at 174, but said it fell behind Fable 5.2 on real-world research tasks, The Decoder reports. Vals AI ranked Astra 7th and Sol 9th on its index and found weak spots in legal research tests, according to The Decoder.
What about safety?
OpenAI says both models reach the "Critical" threshold for cybersecurity under its Preparedness Framework, as GPT-6 Astra did, and that it is adding stricter safeguards for risky cybersecurity requests, The Decoder reports. OpenAI says those safeguards will also limit some legitimate security research.
OpenAI also says Astra is the first of its publicly available models to reach its "High" threshold for AI self-improvement, meaning it can meaningfully speed up AI research, according to The Decoder. OpenAI describes this as a risk and points to tighter controls in research environments and expanded monitoring as countermeasures.
Disclosure: NextFish's AI drafting desk uses Claude, a model made by Anthropic.
Questions readers are asking
What are GPT-6.2 Astra and Sol?
Two new OpenAI models released on Nov. 17. Astra is aimed at demanding professional and scientific work; Sol is a faster, cheaper model for everyday tasks.
How much does GPT-6.2 Astra cost?
OpenAI prices it at $8 per million input tokens and $40 per million output tokens in the API, 20% less than its predecessor, according to The Decoder.
Where can I use them?
In ChatGPT Work and Codex on the Plus, Pro, Enterprise and Edu plans, through the OpenAI API, and in GitHub Copilot for Pro+, Max, Business and Enterprise users, with a gradual rollout.
Sources
- #openai
- #gpt-6-2
- #llm
- #ai-models
- #benchmarks
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
Policy & Safety3 min
California sues OpenAI over Hugging Face hack and AI agent safety
Rob Bonta says OpenAI tested dangerous models without proper oversight. OpenAI says the suit appears to single it out over a challenge all major AI labs have identified.
Chips & Infra3 min
MLPerf Training v6.1 adds post-training test as Rubin, TPUv7 debut
MLCommons says 24 organizations submitted 250 results. Nvidia claims the fastest times, Lambda reports its own gains, and the first Rubin and Google TPUv7 entries appear.
AI3 min
Alibaba launches its Qwen4-Max flagship model, keeps its weights closed
The 2.4-trillion-parameter model takes up to a million tokens and costs $1.50 per million input tokens abroad. Alibaba says smaller Qwen4 models will get open weights.
Robotics3 min
OpenAI folds its robotics team into its main research and applied groups
OpenAI says robotics work continues and that models like GPT-6 Astra can now handle robot tasks directly. Outside experts disagree on what that means for specialised robot models.