OpenAI launched its GPT-6.2 Astra and Sol AI models on November 17, 2026, cutting Astra's API token pricing by 20% to $8 input and $40 output per million tokens. Despite strong benchmark gains—including Astra taking second on the Artificial Analysis Intelligence Index and top rank in 3D design—early AI power users like Matt Shumer and Theo Browne heavily criticized Astra as uneven and difficult to steer in creative and coding tasks.
Model launch and subscriber availability
- ▪GPT-6.2 Astra is available in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu subscribers, as well as via the OpenAI API as gpt-6.2-astra
- ▪Pro-200 subscribers in ChatGPT Work and Codex gained access to Ultrafast mode for the first time starting with GPT-6.2 Sol
- ▪OpenAI released its new frontier language models, GPT-6.2 Astra and GPT-6.2 Sol, on November 17, 2026
Model features and target tasks
- ▪OpenAI said GPT-6.2 Astra is built for complex professional tasks such as law, finance, and accounting, offering enhanced template adherence for slides, spreadsheets, and documents
- ▪OpenAI positioned GPT-6.2 Sol as its efficient, everyday model for routine coding, writing, and multi-step workflows
- ▪GPT-6.2 Astra and GPT-6.2 Sol feature a 1.05 million-token context window and support up to 128,000 max output tokens
API pricing structure and tiers
- ▪For requests exceeding 272,000 input tokens, OpenAI doubles input and cache rates while multiplying output costs by 1.5 for both GPT-6.2 Astra and Sol
- ▪OpenAI set cached input pricing for GPT-6.2 Astra at $0.40 per million tokens, representing a 60% reduction compared to its predecessor GPT-6 Astra
- ▪GPT-6.2 Sol maintained GPT-6.1 Sol API pricing at $2.00 per million input tokens, $0.10 for cached input, and $10.00 per million output tokens
- ▪OpenAI cut standard API token pricing for GPT-6.2 Astra by 20% compared to GPT-6 Astra, charging $8.00 per million input tokens and $40.00 per million output tokens
User and expert criticism
- ▪HyperWrite AI former CEO Matt Shumer stated on X that GPT-6.2 Astra and Sol were essentially unusable during pre-launch testing
- ▪Every CEO Dan Shipper noted that GPT-6.2 Astra often stopped early on end-to-end software builds and exhibited a noticeable house style in writing
- ▪Developer Theo Browne criticized GPT-6.2 Astra as the worst model OpenAI has ever released, citing difficulties with delegated work, creative task steering, and long projects
- ▪Wharton professor Ethan Mollick described GPT-6.2 Astra as very uneven and weird, noting it was harder to steer than rival models despite building a 3D space elevator demo
Benchmark performance and evaluation results
- ▪Artificial Analysis recorded high hallucination rates on the AA-Omniscience benchmark, with GPT-6.2 Astra guessing on 59% of factual questions and GPT-6.2 Sol on 55%
- ▪OpenAI said that on Agents' Last Exam v1.1, GPT-6.2 Astra scored 5.2 points above GPT-6 Astra, reaching 78.3% at less than one-third of the cost
- ▪Artificial Analysis reported that coding agent scores flatlined for GPT-6.2 Astra and dropped 3 index points for GPT-6.2 Sol on the Coding Agent Index
- ▪GPT-6.2 Astra scored 62 points on the Artificial Analysis Intelligence Index, placing second behind Anthropic's Claude Fable 5.2
- ▪GPT-6.2 Astra achieved top rank on the 3D design benchmark AA-3D with 1,318 Elo points, outperforming Claude Fable 5.2
Competitor models and market positioning
- ▪Google opened access to Gemini 4 Carbon a week prior to OpenAI's launch, offering introductory rates of $2 per million input and $10 per million output tokens
- ▪Anthropic launched Claude Fable 5.2 eight days prior to OpenAI's release, matching GPT-6.2 Astra's $8 per million input and $40 per million output rates
- ▪Artificial Analysis noted that OpenAI dominates the Pareto frontier on cost versus performance, with Anthropic and Meta being the only other providers on that frontier
Story comments
Loading comments…