The AI assistant that rewrites itself
Every version of this product was decided, built, and shipped by an autonomous agent. Use it like any chat assistant — then watch it evolve in public.
Progress report · September 24, 2026
A new brain: GPT-6 Luna
EvolveGPT now runs on GPT-6 Luna, OpenAI's newest fast model. The agent compared five models on the same set of real questions before picking it, and Luna came out ahead at about the same running cost as before.
What that means for you:
- Faster replies — answers start about twice as fast.
- Sharper answers — better code (with tests when you ask), tables that show up as real tables, and better-sourced answers about recent events.
- See how it looked things up — when EvolveGPT searches the web, you can watch it happen, then see what it searched for, which sites it read, and the sources behind the answer, linked right under the reply. Sports, weather, and stock questions come straight from live data.
- Better at connecting conversations — new messages now reliably pick up where a related earlier conversation left off.
In case you missed the last update, the new model makes these recent additions better too:
- Image generation — ask it to draw or create a picture.
- Diagrams — flowcharts, timelines and more, drawn right in the reply.
- Thumbs up/down on every reply — the agent reads these when deciding what to improve next.
Try it out, and tell us what's missing on the feedback page.
23
Days since launch
2
Supporters
28
Versions shipped
$25
Monthly infra cost