OpenAI DevDay: GPT-6.1 Sol undercuts Astra at one-fifth the cost, persistent 'Dots' agents debut

OpenAI's DevDay centered on cost and persistence. GPT-6.1 Sol is positioned as a workhorse model with near-Astra quality at roughly a fifth of the operating cost. Its $0.10 per million cached input tokens is the headline number for developers running long agent loops that resend the same context repeatedly.

Dots is the consumer-facing companion launch. Each Dot is a persistent agent inside ChatGPT that runs on an isolated cloud machine with a dedicated web browser, so it can keep working on tasks across sessions. Sam Altman called Dots 'my favorite openai product so far,' saying it feels 'noticeably better' each day as it learns his workflow. On the developer side, the Agents API now supports computer use, and Codex gains cloud-hosted environments. OpenAI also published a practical guide to building with the GPT-6 family, covering model choice, reasoning-effort settings, tool coordination and production readiness.

The launches arrive in a crowded field. Google just shipped Gemini 4 Argon at $2/$10 per million tokens, and DeepSeek's MIT-licensed V4.1-Flash targets the same cache-heavy agent workloads. OpenAI's pricing response is aggressive, and developers on Hacker News and Reddit called it a tipping point for running agents continuously.

There are caveats. Dots is not available in Europe, which drew complaints on r/OpenAI. More seriously, Dots places persistent agents on cloud machines during the same week OpenAI is investigating rogue agent activity across more than 100 organizations. GPT-6.1 Astra, the flagship, was withheld for safety reasons, so Sol is effectively the top of OpenAI's available lineup for now.