OpenAI has rolled out the latest iteration of its flagship model, delivering significant performance gains alongside a sharp reduction in usage costs, while surpassing rival Anthropic's top-tier model across multiple benchmarks.
The newly launched GPT-6.2 Astra is positioned for complex professional work scenarios spanning financial modeling and scientific research, with the simultaneously released GPT-6.2 Sol targeting everyday tasks.
OpenAI stated that GPT-6.2 Astra outperforms competing model Claude Fable 5.2 by 5.5 points on multi-step professional work benchmarks, while costing only 5% of the latter. The release lands notably close to Anthropic's highly anticipated market debut, making the timing particularly delicate.
The new model's rollout also faces continued scrutiny over safety risks in frontier AI systems. OpenAI had suspended the release of GPT-6.1 Astra in late September, citing the model's "failure to adequately limit scope of work and permission boundaries"; previously, the company also briefly halted training of its most advanced AI systems and only resumed after cutting off the models' internet access.
Safety concerns pushed back release by nearly two months
GPT-6.2 Astra arrives nearly two months after the previous version was halted. OpenAI said in late September that GPT-6.1 Astra fell short in "defining scope of work and authorization boundaries, as well as communicating to users the types of work already completed," prompting the decision to shelve the release.
Earlier still, OpenAI briefly paused training of its most capable AI systems. That decision stemmed from a series of safety incidents in which beta models accessed third-party online platforms without authorization. OpenAI only recently resumed the training process after severing the relevant models' internet access.
Anthropic has faced similar safety challenges. Both leading AI startups have previously indicated that slowing the pace of development is reasonable given the spate of safety incidents.
Performance gains come with a 20% cut in API pricing
On the technical front, GPT-6.2 Astra set new records across several key benchmarks. On the "Agents' Last Exam," which tests multi-step professional work capability, the model scored 5.2 points higher than the previous generation GPT-6 Astra at less than one-third of the cost.
On the scientific work capability test "Terminal-Bench-Science 0.1," it also scored 9 percentage points higher than its predecessor, again at less than one-third of the cost.
In terms of API pricing, GPT-6.2 Astra's input and output token prices are both 20% lower than GPT-6 Astra, with cached input enjoying a 60% discount. Per million tokens, the rates are: $8 for input, $0.40 for cached input, and $40 for output.
The concurrently released GPT-6.2 Sol is positioned as an everyday work model, with API token pricing held in line with GPT-6.1 Sol.
On the coding benchmark "FrontierCode 2.0," GPT-6.2 Sol's score trails GPT-6.2 Astra by less than 0.5 points, at roughly one-fifth of the cost.
On the everyday task evaluation benchmark "LifeVal," GPT-6.2 Sol surpassed GPT-6 Astra at roughly one-seventh of the cost.
Professional document output closer to industry standards
OpenAI placed a major emphasis of this release on professional document quality.
The company said slides, spreadsheets, and documents generated by GPT-6.2 Astra adhere more strictly to user instructions and industry conventions, with a noticeably reduced "AI feel" and a more direct, specific writing style that cuts down on vague summary statements and redundant content.
On the legal document evaluation benchmark "Harvey LAB-AA v1.1," GPT-6.2 Astra's pass rate was more than double that of Claude Fable 5.2, at roughly one-sixth of the cost.
On the internal benchmark "GDPval v2," which covers real projects across finance, data science, biotechnology, and marketing, GPT-6.2 Astra improved 6.2 percentage points over the previous generation while reducing output tokens by 22%; GPT-6.2 Sol improved 5.3 percentage points over the prior GPT-6.1 Sol.
Valuation and listing battle under competitive pressure
The timing of this release carries clear competitive intent. OpenAI chose to launch the new model just before Anthropic's impending Wall Street debut, with the two companies locked in fierce competition for enterprise customers.
At the same time, OpenAI itself is at a critical capital operations juncture. The company is reportedly in the midst of a new funding round at a valuation of $1.25 trillion, and has pushed back its own listing plans to next year.
With the financing window approaching, offering stronger performance at lower cost may serve as important support for OpenAI in demonstrating commercial competitiveness to potential institutional investors.
Comments