AI News/AI News

OpenAI's GPT-6.2 Astra and Sol cut costs, but outside tests show mixed results

OpenAI's new models lower Astra's API price and lead on cost-performance, while independent evaluators flag gaps in coding, legal work and hallucinations.

Key points

  • OpenAI released GPT-6.2 Astra and Sol to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API.
  • Astra API pricing is $8 per million input tokens and $40 per million output tokens, 20 percent below its predecessor. Sol pricing is unchanged.
  • Artificial Analysis ranks Astra second on its Intelligence Index, behind Anthropic's Claude Fable 5.2, and says OpenAI leads on cost versus performance.
  • Third-party tests show weak spots: flat coding scores, a high hallucination rate on hard questions, and mixed legal and real-world results.

Why it matters for marketers: Cheaper frontier-level models can lower the cost of agent workflows, but independent tests and early user reviews suggest teams should benchmark on their own tasks before switching.

OpenAI has released GPT-6.2 Astra, aimed at demanding professional work in areas like science, law and finance, and GPT-6.2 Sol, built for faster everyday tasks. Both are rolling out to paid ChatGPT Work and Codex users and to developers through the API. Astra's API price is 20 percent lower than its predecessor's, while Sol keeps its earlier pricing.

For marketers and content teams, the main story is efficiency. Artificial Analysis places Astra second on its Intelligence Index, just behind Anthropic's Claude Fable 5.2, and says OpenAI holds the best cost-performance position. It says Sol matches the earlier Astra's score at about 15 percent of the cost per task. OpenAI also says Astra produces slides that follow template layouts, colors and typography more closely, which may matter for report and deck production.

The picture is less clean elsewhere. Artificial Analysis says coding scores were flat for Astra and slightly lower for Sol, and it reports high hallucination rates on a deliberately hard factual test. Vals AI found the models trailing on legal benchmarks, and Epoch AI said Astra fell behind Fable 5.2 on real-world research tasks. Latent Space also noted unusually negative feedback from some influencers, while attributing part of it to overreaction.

OpenAI also says Astra is the first externally released model to reach its High threshold for AI self-improvement, and that both models reach its Critical cybersecurity threshold, with stricter safeguards added. Teams planning to switch should test Sol first for routine work, compare costs on their own workflows, and keep human review in place for factual and legal content.

Sources

This brief is our own summary. Read the full reporting at the source.

You read about AI automation. We run it.

DocFluence, built by Z Media, writes, designs, films and publishes for your brand every day, in your voice, from one login.