OpenAI's -latest model aliases arrive on OpenRouter — pin or drift
Auto-updating endpoints point at OpenAI's newest models — Astra, Sol, Terra, Luna at 1M context — handy, but silent upgrades can break reproducibility.
Copy markdown
One endpoint, always the newest version
OpenRouter added gpt-astra-latest, gpt-sol-latest, gpt-terra-latest and gpt-luna-latest on Sept 11 — aliases that auto-route to the freshest build in each OpenAI family. Wire your client to one of them and you inherit every future update without touching your code.
The catch: pin it, or it moves under you
Auto-updating cuts both ways — a silent model swap can land mid-project and shift your outputs, costs and eval scores. For anything you need to reproduce, pin the dated model ID and treat -latest as an opt-in, not the default.
The full lineup, priced on one API
All four route through a single endpoint at ~1.05M context: Astra runs $10/$50 per 1M tokens, Terra $2/$12, and Luna just $0.20/$1.20 for high-volume work. Sol is on a promo at $2/$10 — undercutting Terra on output, so the flagship is briefly cheaper to run than the mid-tier.
Elsewhere: OpenAI's math feud escalates
OpenAI pulled its sponsorship of Caltech's 'Mathathon' as prominent mathematicians pushed back on labs training over their work. A live look at research communities drawing hard lines around what AI can learn from.