gpt-6IMAGE
Rumor: Astra's reasoning budget in ChatGPT is a fraction of its API level
Reports say GPT-6 Astra gets about 832 tokens of reasoning budget via the API but only about 128 in Codex, raising suspicion that ChatGPT serves a reduced version of the model.
Codex Update
@CodexResets1Archived 2026.09.30Originally posted 2026.09.29
Codex Update @CodexResets1Originally posted 2026.09.29
🚨 ChatGPT might be serving a nerfed version of OpenAI's own models. > GPT-6 Astra reportedly gets way more reasoning "juice" via the API than through ChatGPT/Codex > Some report ~832 tokens of reasoning budget via API vs just ~128 in Codex > Same model, wildly different reasoning ceiling > API supports max effort — a tier ChatGPT doesn't fully expose > So API users may be getting the full model while ChatGPT users get a capped version Are we all just using a limited version of these models?
This post quotes another post: x.com/mehdicade/status/2104856852106186810
x.com/codexresets1/status/2104858022170316929Archived 2026.09.30 10:19