GPT-6.2 Astra and Sol - Two new OpenAI models: frontier Astra and efficient Sol
by•
GPT-6.2 Astra and Sol are OpenAI’s new models for professional work. Astra is the frontier model for the hardest projects, like slides, websites, 3D designs and science research. Sol is the efficient everyday model for coding and multistep agent work. Per OpenAI, Astra costs under a third per task of GPT-6 Astra. API pricing per million tokens: Astra $8 in, $40 out; Sol $2 in, $10 out. Live in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu, plus the API.

p/runevr
p/jot-2-0
p/beamfy
p/producthunt
p/google-cloud-platform
p/refs
p/dex-by-exmergo
p/reso
p/ai-context-flow
p/kodlo
p/seshka
p/duckfightclub
p/talanoa-2
p/nudgeforme
p/nexupira
p/many-displays
p/dino-kids
p/murror
p/operum
p/tabstack
Replies
OpenAI just released GPT-6.2 Astra and GPT-6.2 Sol, two new models built for real professional work.
Astra is the frontier model for the hardest projects. OpenAI points to slideshows, websites, 3D designs and scientific research, and says its output reads less like AI. Per OpenAI, it sets new highs on evals across professional work and science at less than a third of the cost per task of GPT-6 Astra.
Sol is the efficient everyday model, built for coding and multistep agent workflows. Per OpenAI, on FrontierCode 2.0 (merge-ready changes to real codebases) it nearly matches Astra at about one-fifth the cost.
Availability and pricing:
Live in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API as gpt-6.2-astra and gpt-6.2-sol
API pricing per million tokens: Astra is $8 input and $40 output, and Sol is $2 input and $10 output
Ultrafast mode is available for Pro 500 subscribers on both models, and Pro 200 gets it for the first time, starting with Sol
Both models launch with tighter safeguards for higher-risk cybersecurity help, so some security research requests may be refused. OpenAI says it’s working on reduced-refusal access for approved defenders.
Good fit if you want Astra for your hardest professional and coding work, or Sol when you need most of that quality at a much lower price. The benchmark and cost numbers are OpenAI’s own, so it’s worth testing on your own tasks.
Learn more: OpenAI’s announcement
@rohanrecommends ChatGPT is the best harness for long runing tasks right now. Claude is superior for shorter tasks IMO. Looking forward to test 6.2 today 🚀
I hear these models are shit. Anyone have better experiences?