Tibo's Day 1 | Subscription access gets about 50% faster generation
The GPT-6 Astra and GPT-6.1 Sol change includes partner tools using ChatGPT sign-in. Faster generation is separate from a larger usage allowance.
On this pageCollapse
On Day 1, Tibo announced roughly 50% faster default generation for GPT-6 Astra and GPT-6.1 Sol when used through a subscription. The change was server-side, with no settings adjustment required; at the time, he expected users to notice it within two hours.
Which access paths are included?
The scope includes OpenAI products and partner tools using Sign in with ChatGPT. The post names OpenCode, Pi, Amp and Devin. The relevant condition is ChatGPT subscription sign-in; separately billed API-key requests follow the API’s own guidance.
How much faster will a whole task be?
In a follow-up, Tibo described generation throughput moving from about 30 TPS to 50 TPS. TPS measures generated tokens per second. A complete coding task also includes reasoning, tool calls, file operations and network waits.
A long task therefore may not finish proportionally sooner. Comparisons are more useful with the same model, similar tasks and the same sign-in method. The roughly 50% figure is the author’s announced improvement; this update did not announce a larger subscription allowance.
Sources & corrections
Published UTC