GPT-6.2 Astra Launches to Power-User Backlash

OpenAI's GPT-6.2 Astra and Sol shipped with rising benchmarks and a 20% price cut, yet early testers called the new flagship essentially unusable.

@sharedot

GPT-6.2 Astra Launches to Power-User Backlash

By @sharedot · · 6 pages

  • AI Frontier
  • Openai
  • Model Launches

OpenAI's GPT-6.2 Astra and Sol shipped with rising benchmarks and a 20% price cut, yet early testers called the new flagship essentially unusable.

OpenAI ships Astra and Sol — and power users revolt

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, November 17, both posting new benchmark highs, with Astra getting a 20% API price cut to $8 per million input and $40 output tokens while Sol keeps GPT-6.1 Sol's $2/$10 rates. The reception was harsh.

Investor and former HyperWrite CEO Matt Shumer called the models, per VentureBeat, 'essentially unusable,' and developer and YouTuber Theo Browne wrote that 6.2 Astra was 'the worst model OpenAI has ever released.'

Both models are available now in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, and via the API as gpt-6.2-astra and gpt-6.2-sol.

Why the reaction is such an upset

The timing compounds the sting. VentureBeat notes the release came eight days after Anthropic launched the well-reviewed Claude Fable 5.2 and Mythos 5.2 at the same $8/$40 rates, and a week after Google opened the well-received Gemini 4 Carbon at Sol's introductory prices.

Shumer said he skipped his customary long-form review entirely. Wharton professor Ethan Mollick called this, per VentureBeat, the first OpenAI frontier release since GPT-4.5 that he did not see as a clear step forward, though he could still imagine 6.2 excelling at particular kinds of complex work.

The evidence from early testers

Every CEO Dan Shipper's team found Astra flat versus its predecessor on coding benchmarks, often stopping early on end-to-end builds and pushing a 'noticeable house style.'

Browne reported Astra spent seven days on his Rust TypeScript compiler rewrite without landing an improvement, producing a version that ran 56% slower, then four more days without a change.

Two independent testers asking Astra to design levels for Outer Wilds and Majora's Mask both got music-and-bells water levels named 'Bellwether.'

On OpenAI's own launch charts, Astra beat GPT-6 Astra on all nine benchmarks and Sol beat 6.1 Sol on eight of nine — but independent Codex testing found a flat result for Astra and a regression for Sol.

What it means for builders

For practitioners, the launch lands amid a market that switches fast and prices by task, not tokens.

SaaStr's analysis of the Databricks AI Index reports single-provider enterprises fell from 64% to 28% in one quarter, and Anthropic's Opus 5.5 captured 37.9% of proprietary tokens in six weeks after launch — evidence that a poorly received flagship can be swapped out with a config change.

It also shows the most capable model is not the most used: Claude Fable 5.1, Databricks' best coding model, saw its share fall from 5.5% to 0.6%.

Not all reaction was negative — raindrop.ai CTO Ben Hylak and AI advisor Ashe Magalhaes praised Astra's animated visuals, motion graphics, VFX, and agentic video editing.

Sources

  1. venturebeat.com › 'Essentially unusable': OpenAI launches GPT-6.2 Astra and Sol to negative reception from AI power users
  2. saastr.com › 5 Interesting Learnings from the Databricks AI Index

More on AI Frontier