GPT-6.2 Arrives With Big Promises and Immediate Pushback

OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, but the launch is already colliding with fierce criticism from the people who tested them first. The company promises better work across code, documents, spreadsheets, slides, gaming, and 3D design; early reactions say Astra can deliver impressive results one moment and stall on basic direction the next.
The two models arrived Tuesday, November 17, 2026, in a release that puts OpenAI under pressure from rivals and power users alike. Astra carries a 20% price cut through the application programming interface, while Sol keeps the pricing of its predecessor, GPT-6.1 Sol.
Astra Promises More Control, But Testers Found Friction
OpenAI positions Astra for demanding assignments such as changing a complex codebase, building a working web application, or producing a document that follows company templates. The company says its updated frontier model creates slides, spreadsheets, and documents that “read less like AI,” follows templates more closely, and performs better in websites, gaming, and 3D design.
Astra costs $8 per million uncached input tokens and $40 per million output tokens. Sol costs $2 per million input tokens and $10 per million output tokens, making it the cheaper choice for routine coding, writing, and workflows that require several steps. Both models are proprietary and hosted.
They are available now in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users. Developers can also access them through the OpenAI API with the model names gpt-6.2-astra and gpt-6.2-sol.
OpenAI says Sol is “smarter and more efficient” than its predecessor, and that many of its own employees already use it as their default model for technical and other work. Astra, however, has faced a much rougher introduction.
Investor and former HyperWrite AI CEO Matt Shumer said he would not publish the longer written review he has customarily released for preceding OpenAI flagship models. In a post on X, Shumer called the new models “essentially unusable,” adding that when he assigned Astra work as a sub-agent, Claude models supervising it redid the work.
Developer and YouTuber Theo Browne delivered an even harsher verdict, writing that 6.2 Astra was “the worst model OpenAI has ever released.” He singled out delegated work and creative tasks, then published a long X post titled “Why I hate GPT-6.2 Astra,” which he said contained the full text of a letter he had sent OpenAI during early testing.
Impressive Demos, Uneven Long Projects
Browne described a test involving his experimental TypeScript compiler rewrite in Rust. He said Astra spent seven days without landing an improvement, then produced a version that ran 56% slower after focusing on reducing its size. When he redirected the model toward speed, Browne said it spent another four days without landing a change.
That account did not make Astra useless in every setting. Browne credited it with catching issues other models missed in code reviews and praised its 3D, game development, and computer-use capabilities. He also said Astra helped him devise a reverse proxy that he expects will save him $25,000 a month.
His main complaint focused on keeping Astra moving through long projects and understanding what it had done afterward. According to Browne, OpenAI addressed a few of his complaints before launch, but he still preferred Claude Opus 5.5 for his development work.
A separate creative test raised questions about whether Astra can produce distinct ideas. Browne and another tester independently asked the model to design new levels for Outer Wilds and Majora’s Mask. Both received water levels centered on music and bells, and both levels carried the same name: “Bellwether.”
Dan Shipper, CEO of generative AI publication Every, said his team spent several weeks testing both models. He found Astra performed about as well as its predecessor on most of Every’s coding benchmarks, but often stopped early on end-to-end builds.
Shipper also found that Astra’s push for brevity could make writing hard to follow or leave out needed explanation. He praised its default design instincts, yet said its outputs were difficult to vary and carried a “noticeable house style.”
Competition Makes the Launch Harder to Ignore
Wharton professor Ethan Mollick had a more mixed experience. Astra built a playable space elevator in Blender and Unity, and Mollick said the demo simulated weightlessness. Yet he described the model as “very uneven & weird,” found it harder to steer than rival models, and said he still reached for other models.
In a LinkedIn comment on his demo post, Mollick called Astra the first OpenAI frontier release since GPT-4.5 that he did not see as a clear step forward. He still said he could imagine 6.2 excelling at particular kinds of complex work, while Raindrop.ai CTO Ben Hylak and AI advisor Ashe Magalhaes shared more positive views of Astra.
The timing adds pressure. Anthropic launched the well-reviewed Claude Fable 5.2 and Claude Mythos 5.2 eight days earlier, and both carry the same $8 and $40 per million standard input and output prices as Astra. Google also opened access to its well-received Gemini 4 Carbon a week earlier at introductory rates equal to Sol’s.
The broader conversation around OpenAI is also pulling in other claims and debates. Nvidia says it does not intend to reinstate Jensen Huang as CEO, while an Ars OpenForum thread started by JournalBot carried the title, “AI is ‘the most likely reason’ you’ll die, OpenAI board member warns.” Another post in that discussion begins, “I have PART of the answer.”
GPT-6.2 now faces a clear test beyond launch-day demos: can Astra turn its strong flashes into dependable progress across long, messy assignments? OpenAI has put forward a cheaper, more capable model family, but users will decide whether those gains outweigh the steering problems, repeated outputs, and stalled projects appearing in the first wave of testing.
Based on
- OpenAI’s newest release lands to rough reviews — therundownai.beehiiv.com
- ‘Essentially unusable’: OpenAI launches GPT-6.2 Astra and Sol to negative reception from AI power users | VentureBeat — venturebeat.com
- OpenAI releases its updated GPT-6.2 Astra model. | The Verge — theverge.com
- AI is “the most likely reason” you’ll die, OpenAI board member warns | Page 7 | Ars OpenForum — arstechnica.com




