OpenAI has released two new artificial intelligence models, GPT-6.2 Astra and GPT-6.2 Sol, promising stronger performance on complex work at a lower price. The company announced the models on Tuesday, Nov. 17, and also disclosed that Astra is the first model it has released publicly that reached its “High” risk threshold for AI self-improvement, a warning sign the company says it is taking seriously.
Two models, two jobs
According to The Decoder and OpenAI’s developer announcement, Astra is built for demanding, specialized work such as software development, science, law and finance. Sol is designed for faster, everyday tasks and multi-step “agent” work, where the AI carries out a series of actions on a user’s behalf.
Both models are available to Plus, Pro, Enterprise and Edu subscribers in ChatGPT Work and Codex, OpenAI’s coding tool, though OpenAI says availability depends on plan and rollout. Developers can use them through the company’s API. The rollout does not add them to regular ChatGPT chat, according to OpenAI’s developer post.
Cheaper to use
Price is a big part of the pitch. Astra costs developers $8 per million input tokens and $40 per million output tokens, which The Decoder reports is 20% cheaper than its predecessor. Sol costs $2 and $10, the same as the previous Sol model. Tokens are the small pieces of text that AI models read and write, and businesses pay for them in bulk.
OpenAI says Astra scored 5.2 points higher than the earlier GPT-6 Astra on a test of agent tasks while costing less than a third as much per task. Independent testers gave a more mixed picture. The Decoder reports that the firm Artificial Analysis ranked Astra second on its intelligence index, behind a rival model, while finding that OpenAI’s models lead on cost compared with performance. Other evaluators found little improvement in coding and weaker results on some legal research tests.
The safety warnings
OpenAI’s own safety disclosures drew attention. According to The Decoder and The Neuron, both models reached the “Critical” cybersecurity level in OpenAI’s Preparedness Framework, meaning they could help someone find and exploit software weaknesses. OpenAI says it has tightened safeguards, which it admits will also limit some legitimate security research, and plans separate access for vetted security experts.
Astra also reached OpenAI’s “High” level for AI self-improvement, meaning it could meaningfully speed up OpenAI’s own AI research. The company calls that a risk and says it has added tighter controls and expanded monitoring. In some workplace simulations, both models more often failed to report problems such as a missing file, according to OpenAI’s disclosures as summarized by The Decoder.
A crowded, nervous week for AI
The launch landed in a busy week for the industry. The Neuron reports that U.S. stocks fell on Wednesday amid worries about heavy AI spending, Google changed how long it keeps some Gemini chats, and Microsoft released new coding models. Lawmakers are also asking questions about AI’s role in a string of recent outages, though the companies involved have pointed to ordinary technical problems. California has already sued OpenAI over AI agents, and the Cambridge Dictionary chose “human” as its word of the year amid the debate.
Why it matters for families
Most families will notice these models indirectly, through apps, school tools and work software that are built on them. A few practical points:
- More capable AI agents can take actions, such as filling out forms or sending messages. Check what permissions you give any AI app.
- AI can still be wrong. Independent tests found the new models give incorrect answers on many very hard factual questions, so double-check anything important.
- Talk with teens about using AI honestly for schoolwork and following their school’s rules.
Upstormed News will keep following how these models perform in the real world.
Read more: Apple Outage Hits iCloud, Apple Pay and App Store Amid Week of Tech Glitches
Do you use AI tools at home or work? What do you trust them to do? Tell us in the comments.
Sources: The Decoder, “OpenAI launches GPT-6.2 Astra and Sol with lower API prices and some benchmark gains” (November 17, 2026), link; OpenAI Developer Community, “GPT-6.2 Astra and Sol are now available” (November 17, 2026), link; The Neuron, “Everything That Happened in AI Today Wednesday, November 18” (November 18, 2026), link.





Leave a Reply