N環球 N GLOBE
US Technology

OpenAI Releases GPT-6.2 Astra and Sol; Flagship 20% Cheaper, First High Self-Improvement Rating

· N GLOBE

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, available now in ChatGPT Work, Codex and via API; Astra costs 20% less than its predecessor, both models are rated Critical for cybersecurity, and Astra is the first externally released model rated High for AI self-improvement.

OpenAI released its next-generation models GPT-6.2 Astra and GPT-6.2 Sol on November 17, available now to Plus, Pro, Enterprise and Education users in ChatGPT Work and Codex, and to developers through the API. Astra is the flagship, aimed at the hardest professional projects, while Sol is the efficient model for everyday work. The two models come only to ChatGPT Work and Codex; regular Chat still uses GPT-6.

On price, Astra's standard API rates are 20% lower than its predecessor: $8 per million input tokens and $40 per million output tokens, with cached input at $0.40. Sol keeps GPT-6.1 Sol's rates of $2 input and $10 output per million tokens. Both models support Ultrafast mode; Pro 500 subscribers get it with either model, while Pro 200 subscribers get it first with Sol.

Benchmark gains

According to OpenAI's own evaluations, Astra improves 6.2 points over GPT-6 Astra on GDPval v2 while using 22% fewer output tokens, and gains 5.2 points on Agents' Last Exam v1.1 at less than one-third of the cost. On LifeVal, which measures challenging everyday tasks, Sol outscores GPT-6 Astra at about one-seventh of the cost. These are OpenAI's own figures and have not been independently verified.

Safety ratings draw attention

The system card rates both models Critical for cybersecurity and High for biological and chemical capabilities under OpenAI's Preparedness Framework; Astra is also rated High for AI self-improvement, the first externally released OpenAI model to reach that level. OpenAI said the capability gains bring tighter default restrictions on higher-risk cybersecurity assistance, acknowledging that the limits can restrict legitimate vulnerability research and security testing.

GPT-6.2 arrives about three weeks after GPT-6.1, continuing OpenAI's faster model cadence this year. It also comes a day after California sued OpenAI over "rogue agents"; the new models' safety ratings and tightened safeguards land as scrutiny of AI safety intensifies.

(Sources: OpenAI; Unite.AI. Image: OpenAI logo file photo.)

💬 Comments

No comments yet. Be the first!