OpenAI unveils GPT-6.2 Astra and Sol amid mixed early feedback
Միայն փաստեր

OpenAI unveils GPT-6.2 Astra and Sol amid mixed early feedback

Ամփոփում

OpenAI launched the GPT-6.2 Astra and Sol models with lower pricing for Astra and unchanged rates for Sol, but several AI power users reported usability issues and uneven performance.

OpenAI introduced two new models, GPT-6.2 Astra and GPT-6.2 Sol, on Tuesday. Astra is priced at $8 per million uncached input tokens and $40 per million output tokens, a 20% reduction from its predecessor, while Sol retains the $2 input and $10 output rates of GPT-6.1 Sol. Both models are available through ChatGPT Work, Codex, and the OpenAI API under the identifiers gpt-6.2-astra and gpt-6.2-sol.

Early testers expressed divergent views. Investor Matt Shumer described Astra as "essentially unusable" in a post on X, and developer Theo Browne called it "the worst model OpenAI has ever released," citing difficulties with delegated work and creative tasks. Wharton professor Ethan Mollick reported that Astra could generate a functional space-elevator demo in Blender and Unity but found its behavior "very uneven & weird" and harder to control than competing models.

Dan Shipper, CEO of the generative-AI outlet Every, said Astra performed similarly to its predecessor on most coding benchmarks but often halted early on end-to-end builds and produced overly brief writing. He noted a consistent "house style" that could limit flexibility for teams needing varied tones. Browne highlighted repetitive design outputs, such as generating identical water-level concepts for different game projects, and described prolonged attempts to improve TypeScript code that resulted in slower performance.

Other users, including Raindrop.ai CTO Ben Hylak and AI advisor Ashe Magalhaes, offered more positive remarks, praising Astra’s capabilities in animated visuals, motion graphics, VFX, and agentic video editing. OpenAI indicated that many of its own employees already use Sol as the default model for technical tasks.

Pricing comparisons show Astra costing four times as much as Sol per token, matching Anthropic's Claude Fable 5.2 rates but higher than Anthropic's cache-read fees. Google’s Gemini 4 Carbon is offered at rates comparable to Sol. Both new OpenAI models support a 1.05 million-token context window and up to 128,000 output tokens, with higher charges for requests exceeding 272,000 input tokens.

In benchmark tables released by OpenAI, GPT-6.2 Astra outperformed GPT-6 Astra on all nine head-to-head tests, and GPT-6.2 Sol improved on eight of nine versus GPT-6.1 Sol. Independent testing by Codex showed flat results for Astra and a regression for Sol on broader coding-agent evaluations. OpenAI’s system card reported mixed outcomes, with improvements on some health-benchmark tasks but increased instances of undisclosed obstacles in simulated workplace scenarios.

The reception underscores the competitive landscape, as Anthropic recently launched Claude Fable 5.2 and Mythos 5.2, and Google opened access to Gemini 4 Carbon with introductory pricing similar to Sol. Whether Astra’s lower per-token cost and Sol’s unchanged rates translate into reduced overall project expenses will depend on the models’ ability to complete tasks accurately and efficiently.

Աղբյուր

VentureBeat

Շարունակեք կարդալ — անվճար

Ստեղծեք անվճար հաշիվ՝ նորություններին հետևելու համար։ Քարտ չի պահանջվում։