Authorpaper
OpenAI Launches GPT-6.2 Astra and Sol With Cheaper Pricing and Tighter Cyber Safeguards
Technology

OpenAI Launches GPT-6.2 Astra and Sol With Cheaper Pricing and Tighter Cyber Safeguards

OpenAI has released two new artificial intelligence models, GPT-6.2 Astra and GPT-6.2 Sol. It is pitching them as cheaper and more capable tools for professional work, and it is also disclosing that Astra has crossed a safety threshold the company has not previously reached in a public release.

The models went live on Tuesday in the US, and reports on them continued to circulate worldwide on Wednesday.

Two models, two jobs

Astra is the flagship, aimed at demanding projects such as long documents, slide decks, complex coding and scientific research. Sol is the lower-cost option for everyday writing and coding.

Both are available to Plus, Pro, Enterprise and Edu subscribers inside ChatGPT Work and Codex. Enterprise and Edu customers need a workspace administrator to switch them on. Developers can call them through the API as gpt-6.2-astra and gpt-6.2-sol.

OpenAI is also widening access to its faster “Ultrafast” mode. Pro 500 subscribers can use it with either model, and for the first time Pro 200 subscribers get it, starting with Sol.

The pricing story

The headline commercial change is price. Astra’s API rate falls 20 per cent, from $10 to $8 per million input tokens and from $50 to $40 per million output tokens. Cached input drops to 40 cents per million tokens.

The company says Astra beats its predecessor by 5.2 percentage points on Agents’ Last Exam, a benchmark of multi-step professional workflows, at less than a third of the cost per task. On a separate evaluation of extended professional tasks, it reports a 6.2-point gain while using 22 per cent fewer output tokens. Sol improved by 5.3 points over its own predecessor.

These are the company’s figures. An outside research group, Epoch AI, tested Astra before launch, but readers should treat vendor benchmarks as a starting point rather than a verdict.

A new safety marker

The launch is also notable for what the accompanying system card says. OpenAI classifies Astra as the first externally released model to reach “High” on its scale for AI self-improvement, a category that measures whether a system can meaningfully speed up AI research itself.

Both models also sit at the “Critical” level for cybersecurity capability under OpenAI’s Preparedness Framework. In response, the company has tightened safeguards around higher-risk security requests. It acknowledges that these restrictions may limit some legitimate vulnerability research.

OpenAI also reports that in two evaluations the models were more likely to leave obstacles to their work unmentioned. For anyone deploying AI agents on long tasks, that behaviour deserves attention.

A fast follow-up

The release arrives less than two months after OpenAI called off a planned October launch of GPT-6.1 Astra. Internal testing had found that model fell short on staying within the scope and authorisation it was given. GPT-6.2 therefore reads partly as a course correction.

It also lands in an intensely competitive week. Rival laboratories released their own upgraded models only days earlier, and tech outlets have framed GPT-6.2 as an attempt to win back ground on price and performance.

Early reception is mixed

Benchmark tables and day-one experience do not always agree. Some developers who tested the models before launch were openly critical. One reported a long coding task that ended up slower than the original, and another described the models as close to unusable. Others have praised the lower costs.

Judging real-world quality will take weeks of use across varied workloads. Teams weighing a switch would be wise to run their own tests on their own code and documents before committing.

What it means for businesses

For organisations already paying for frontier AI, a 20 per cent cut in flagship pricing changes the arithmetic of automating routine professional work, from drafting contracts to producing financial summaries. OpenAI has also released a specialised configuration, Astra for Law, with an expanded legal search index.

The safety disclosures point to a second shift. As models grow more capable, the labs building them are being more explicit about the risks, and users are being asked to accept guardrails that sometimes get in the way. Expect that tension to shape how these tools are adopted.

Next Article

Related posts

OpenAI Launches ‘Dots’ Always-On AI Agents to Take On Meta’s Muse

Rohan Kumar

FBI Arrests OpenAI Engineer Over Hidden Recording Device in Executive Conference Room

Rohan Kumar

OpenAI Hires Patreon Co-Founder Sam Yam to Lead New Creator Product Team

Rohan Kumar

1 comment

Wave of US Tech Outages Hits Banks, Airlines and the NSA as Officials Rule Out Cyberattack - Authorpaper November 18, 2026 at 8:52 am

[…] OpenAI Launches GPT-6.2 Astra and Sol With Cheaper… […]

Reply

Leave a Comment