
Share
A 20% price cut wasn't enough to offset harsh early reviews of GPT-6.2 Astra and Sol. With Anthropic and Google both shipping well-received rivals in the same two weeks, the timing couldn't be worse for OpenAI.
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, touting new benchmark highs and a 20% price cut on Astra's API rates. The market response from the people who matter most, the developers and power users who stress-test these models before the broader public ever touches them, was blunt and largely negative.
Investor and former HyperWrite AI CEO Matt Shumer said the models "were essentially unusable" and declined to publish the longer review he has written for past OpenAI flagships. Developer and YouTuber Theo Browne went further, calling Astra "the worst model OpenAI has ever released." Both pointed to failures in delegated work and creative output, the kinds of multi-step, open-ended tasks that matter most to professional users willing to pay premium rates.
Not every review was damning. Wharton professor Ethan Mollick built a playable space elevator in Blender and Unity using Astra, simulating weightlessness in the process. But he described the model as "very uneven & weird" and harder to steer than its rivals. On LinkedIn, he called it the first OpenAI frontier release since GPT-4.5 that he did not see as a clear step forward, while still allowing that 6.2 might excel at specific kinds of complex work.
Dan Shipper, CEO of the generative AI publication Every, offered the most measured account after several weeks of testing. Astra matched its predecessor on most of Every's coding benchmarks but often stopped short on end-to-end builds. Its push for brevity, he found, could strip out explanation readers needed. He praised its design instincts but flagged a "noticeable house style" that made outputs hard to vary, a real limitation for teams serving multiple brands or audiences.
Browne's complaints ran deeper still. In a lengthy post titled "Why I hate GPT-6.2 Astra," reproduced from a letter he sent OpenAI during early testing, he detailed a seven-day stretch where Astra attempted to rewrite his experimental TypeScript compiler in Rust and produced a version running 56% slower. Redirected to prioritize speed, it spent another four days without landing a usable change. He did credit Astra for catching code review issues other models missed and for strong 3D, game development, and computer-use capabilities, including a reverse proxy he expects will save $25,000 a month. His core criticism was about momentum: keeping Astra moving on long projects and understanding what it had actually done afterward.
Browne also demonstrated a quirk that undercuts OpenAI's creative-use pitch for Astra. He and another tester independently asked the model to design new levels for two different games, Outer Wilds and Majora's Mask. Both got water levels built around music and bells, and both got the same name: "Bellwether." Two unrelated prompts, same creative output, same title. That's not a coincidence testers want to see in a model being marketed for creative and coding work alike.
Shumer reported a similar pattern when he used Astra as a sub-agent: the Claude models supervising it ended up redoing the work. An X user posting as Jacob echoed the dissatisfaction and speculated it traced back to how OpenAI trained the models to behave. Not everyone agreed. Raindrop.ai CTO Ben Hylak and AI advisor Ashe Magalhaes both praised Astra's strengths in animated visuals, motion graphics, VFX, and agentic AI video editing. OpenAI, for its part, says many of its own employees already default to Sol for technical work.

The numbers matter here because OpenAI is asking buyers to pay flagship rates for a model that several experienced testers say underperforms its predecessor on real work. Astra now costs $8 per million uncached input tokens and $40 per million output tokens at standard rates, down 20% from GPT-6 Astra's $10 and $50. That cut carries through the long-context and Fast tiers too. Sol holds steady at $2 input and $10 output, identical to its predecessor across every tier.
Those Astra rates put it exactly level with Anthropic's Claude Fable 5.2 and Mythos 5.2, launched eight days earlier to much better reviews at the same $8 and $40 price points. Anthropic undercuts OpenAI on cache economics, though: it charges half as much for cache reads on its flagship models, and $10 per million tokens for a five-minute Fable cache versus OpenAI's $10 for just a 30-minute Astra cache. Google's Gemini 4 Carbon, opened a week before OpenAI's launch to its own positive reception, matches Sol's $2 and $10 introductory rates before stepping up to Opus-level pricing on December 1.
Sitting among dozens of competing models on a broader market chart, Sol lands squarely mid-pack at $12 per million combined tokens for short-context standard use, cheaper than Grok 4.7, Qwen3.8-Max, and Gemini 4 Hydrogen, but far pricier than budget options like Meta's Muse Ember or DeepSeek's Flash variants. Astra sits near the top of the chart at $48 combined, tied with Anthropic's new flagships and well below the older GPT-6 Astra's $60, but still a 4x premium over Sol for identical token counts.
OpenAI cut prices and lifted benchmark scores, but neither moved the needle with the testers who shape enterprise and developer sentiment ahead of wider adoption. The complaints center on reliability over long tasks, inconsistent creative output, and a house style that resists customization, not raw capability. That distinction matters for OpenAI's near-term competitive position. Anthropic and Google both shipped comparably priced, better-received models in the same two-week window, and Anthropic is doing so just ahead of a Nasdaq listing with $16.3 billion in first-half 2026 revenue behind it. Investors watching the AI infrastructure race should treat launch-day pricing as necessary but not sufficient. Benchmark wins and discounts mean little if the people paying premium rates for a flagship model find themselves reaching for a competitor's tool to finish the job.
Tags
Original Sources
'Essentially unusable': OpenAI launches GPT-6.2 Astra and Sol to negative reception from AI power users
↗ https://venturebeat.com/technology/essentially-unusable-openai-launches-gpt-6-2-astra-and-sol-to-negative-reception-from-ai-power-users
About the author
Marcus began tracking AI's market implications in 2016, noticing AI-related patent filings accelerating ahead of earnings upgrades before most of the sell-side had caught on. A former fixed-income quantitative analyst, he spent two decades building models that priced risk across emerging markets before pivoting to cover the economic impact of AI full-time. His writing translates opaque technical developments into clear risk/reward terms — and he's rarely diplomatic about the gap between AI valuations and underlying fundamentals. He believes most market participants still underestimate AI's long-run deflationary effect on knowledge work.
More from The Analyst →This Week's Edition
18 November 2026
29 articles
Related Articles

Elsevier Gives ClinicalKey AI Full EHR Access, Betting on Patient Context Over Chatbots
Products & Applications · 6 min

BMO Claims Quantum Utility Milestone in AI Feature Selection, Using 100x Less Data
Products & Applications · 5 min

National Academies Report Calls for Faster, Tougher Retraction Standards
Models & Research · 5 min
Related Articles

Elsevier Gives ClinicalKey AI Full EHR Access, Betting on Patient Context Over Chatbots
Products & Applications · 6 min

BMO Claims Quantum Utility Milestone in AI Feature Selection, Using 100x Less Data
Products & Applications · 5 min

National Academies Report Calls for Faster, Tougher Retraction Standards
Models & Research · 5 min
More Stories
© 2026 Cedar & Bloom. All rights reserved.