Microphone Review

Audio and tech news for people who like good sound

Brought to you with help from

OpenAI ships GPT-6.2 Astra and Sol with cheaper API rates

OpenAI ships GPT-6.2 Astra and Sol with cheaper API rates

OpenAI has launched two new models, GPT-6.2 Astra and GPT-6.2 Sol. The company says both give better results with fewer output tokens, which stretches a paid plan further.

Two models, two jobs

Astra is the frontier model, aimed at the hardest professional projects. OpenAI says the decks, tables and written deliverables Astra produces feel less machine-made, and that they stick closer to supplied templates and field conventions. It also writes cleaner code and builds better frontends, especially for websites, gaming and 3D design.

Sol is the efficient model for everyday work. Compared with GPT-6.1 Sol, OpenAI calls it smarter and more efficient, and says it lets people finish more tasks on the same paid plan. On Cognition's FrontierCode 2.0, Sol lands within 0.5 points of Astra at about one-fifth the cost.

What the benchmarks say

On Agents' Last Exam, which tests multistep professional work, Astra scores 5.2 points above GPT-6 Astra at less than a third of the cost. It is 5.5 points ahead of Claude Fable 5.2 at 5% of the cost.

On Harvey LAB-AA v1.1, a legal-assignment test, Astra more than doubles the pass rate of Claude Fable 5.2 at about one-sixth the cost. On OpenAI's own GDPval v2, Astra is up 6.2 percentage points on GPT-6 Astra while using 22% fewer output tokens. In Artificial Analysis's AA-3D, human judges ranked Astra first at 1,318 Elo, ahead of Claude Fable 5.2 at 1,271 and Sol at 1,259.

Pricing and access

Astra's API prices are 20% lower than GPT-6 Astra's. Cached input gets a 60% discount. The rates per million tokens are $8 for input, $0.40 for cached input and $40 for output. Sol keeps GPT-6.1 Sol's rates of $2, $0.10 and $10.

The models are available now to Plus and Pro users, and to Enterprise and Edu users once their workspace admin enables it, in ChatGPT Work and Codex. Pro 500 subscribers can use Ultrafast with either model. For the first time, Pro 200 subscribers get Ultrafast too, starting with Sol. Developers can call gpt-6.2-astra and gpt-6.2-sol through the API.

Safeguards and a known quirk

Both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities under OpenAI's Preparedness Framework. The company launched them with tighter safeguards for higher-risk cybersecurity requests. It says these limits also also get in the way of some lawful security research and penetration work.

Astra is also OpenAI's first externally released model to reach the High threshold for AI self-improvement. OpenAI describes that as equal to adding a skilled research engineer to help each OpenAI researcher. To manage the risk, it built Control Room, a tool that monitors research workloads and can stop a model's output when intervention is needed.

In two alignment tests, the models were more likely to fail to mention blockers on their tasks, for example a missing file or no search tool. OpenAI added Codex instructions that substantially reduce this and says it is still investigating the behavior.

Source: OpenAI

Find a story

What is this place

Audio and gadget news with context, collected in one tidy archive.
Ping the team: [email protected] © Microphone Review. Every right kept.