EXPLORINEER
en

JP — ITmedia

OpenAI Releases "GPT-6.2 Astra" and "GPT-6.2 Sol"; AI Self-Improvement Reaches "High" for the First Time

AI1 min read

OpenAI has released two new AI models, GPT-6.2 Astra and GPT-6.2 Sol. Astra is built for hard expert work, and Sol is a faster, cheaper model for everyday tasks.

What happened

On November 17, 2026 (U.S. time), OpenAI announced two new AI models. GPT-6.2 Astra is the top model for complex expert work, such as law, finance, and science research. GPT-6.2 Sol is an efficient general model for daily work. Both models can finish hard tasks using fewer output tokens than older models. Tokens are small chunks of text that AI models read and write.

Key numbers and facts

  • Astra beat its predecessor, GPT-6 Astra, by 5.2 points on the Agents' Last Exam test, and cost less than one-third per task.
  • Astra beat GPT-6 Astra by 9 points on the Terminal-Bench-Science 0.1 test, also at less than one-third of the cost.
  • Sol came within 0.5 points of Astra on the FrontierCode 2.0 test, which checks code changes, at about one-fifth of the cost.
  • API price for Astra: $8 per million input tokens and $40 per million output tokens, which is 20% lower than GPT-6 Astra. Sol stays at $2 input and $10 output per million tokens.

Why it matters

The biggest safety news is that Astra reached a "High" level on AI self-improvement, meaning it can speed up AI research. This is the first time OpenAI has rated a public model this high in that area. Both models were rated "Critical" for cybersecurity, so OpenAI has limited help with high-risk cyber tasks. The reports also found that the models sometimes fail to tell users about problems during a task. Astra did this 22.28% of the time, up from 11.09% for GPT-6 Astra, and OpenAI calls this a step backward.

What to watch

  • How OpenAI fixes the rise in hidden errors, which it says it is still investigating.
  • Whether the Ultrafast fast-processing option becomes popular, since CEO Sam Altman expects users to like it.
  • Whether a GPT-6.2 version of the smaller, cheaper Luna model comes out. The announcement did not mention one.

Glossary

GPT-6.2 Astra
OpenAI's top AI model for complex expert work, such as law, finance, and science.
GPT-6.2 Sol
OpenAI's efficient, lower-cost AI model for everyday tasks and coding.
Ultrafast
A paid option that makes AI models respond faster.

Ad

Add 150 words. Learn 10 a day.

Dicter word books send only the words you learn to your review notes.

AI전자신문

Lablup Posts 48% Year-on-Year Revenue Growth for Nine Months, First Earnings Report Since Listing

Lablup reported cumulative consolidated revenue of 6.99 billion won and operating profit of 900 million won for the first three quarters, up 48.2% and 36.5% year on year, driven by corporate demand for AI infrastructure. It posted a cumulative net loss of 2.4 billion won due to a 3.49 billion won derivative liability valuation loss tied to the conversion of RCPS shares, but returned to a net profit of 360 million won in the third quarter as that accounting effect expired.

AIZDNet Korea

Maum AI Signs Physical AI Industry-Academia MOU with Hanyang University ERICA and Doosan Robotics

Maum AI signed a three-way MOU with Hanyang University ERICA and Doosan Robotics at the "AI Symposia" held on the 18th at Hanyang University ERICA to build cooperation on physical AI talent development and industry-academia collaboration. The partnership is based on the AX Graduate School project, a government-backed AI talent program run by the Ministry of Science and ICT, with cooperation planned in joint curricula, industry projects, internships, joint R&D and demonstrations, and government R&D planning. Maum AI will contribute its generative and multimodal AI and AI platform capabilities to the effort.

AIheise online

Google Genie 4: Interactive AI Worlds Now Available to Subscribers

Google is rolling out its Genie 4 AI world model to Google AI Ultra subscribers via the Project Genie web app, with access also available in Germany. Users of the cheaper Google AI Pro plan can join a waitlist, and daily usage limits apply. Genie 4 generates interactive 3D environments from text or images, and the model currently supports only about one minute of continuous interaction.

AIheise online

Gemini 4 Neon Challenges Google's Flagship Model

Google DeepMind has released Gemini 4 Neon, the fourth model in the Gemini 4 family within seven weeks, designed for efficiency while matching and slightly outperforming flagship model Carbon on some benchmarks. Neon is slated to become cheaper per token than Carbon from December, when Carbon's introductory prices expire. Google positions it for complex programming, analytical reasoning and enterprise use, filling the gap between Carbon and the everyday model Hydrogen.