ChatGPT users report silent downgrade from GPT-5.6 to 5.5-mini
From August 23 to 24, multiple ChatGPT users reported the same issue: after selecting the GPT-5.6 Sol model in the interface (especially in Medium/High modes), the actual responses were generated by gpt-5.5-mini, with no downgrade notice or quota warning from the system, and OpenAI has yet to respond.
Confirmed
- @vvebevv reported that multiple Plus users encountered this issue; network captures show requests were directed to 5.6 Sol, but the server responses came from gpt-5.5-mini
- @devilozz noted that the assistant's self-introduction also shows 5.5-mini, and the downgrade behavior was never disclosed to users, raising transparency concerns
- @ThutmoseCYH relayed Reddit user feedback that server metadata showed routing to GPT-5.5-mini; the issue has persisted for days, affecting Plus and Pro users, with symptoms including extremely fast responses, no reasoning, and low quality
- @coylter, through a 100% reproducible test, found that "Instant" mode still calls 5.6 Sol, while "Medium," "High," and "Extra High" modes all fall back to 5.5 mini, and noted that 5.5 mini's performance has noticeably declined
- @Pinksmurf04 reported another individual anomaly: a Project bot organizing Japanese exam questions thought for only a few seconds before parroting back the user's prompt instead of performing the task; when asked about its identity, it claimed in English to be "GPT 5.6 Sol" and gave other absurd answers in Chinese
Unconfirmed
- Whether OpenAI intentionally adjusted its routing strategy (e.g., for cost control) or it is a technical failure remains unclear with no official statement
- Whether @Pinksmurf04's parroting and false identity anomaly is directly related to the downgraded routing is based on a single report only
Why it matters
- If model selection is silently swapped, the capabilities users pay for in their subscriptions don't match the actual experience, directly undermining trust
- Multiple users corroborated the issue with packet captures, metadata, and reproducible tests, showing this is not an isolated perception but a verifiable change in routing behavior
- The incident has sparked widespread questions about model service transparency and users' right to know
2026-08-23 ~ 2026-08-24 · 4 related posts
Primary sources
- [source] User finds ChatGPT routing changed: Medium/High effort tasks now default to 5.5 mini — coylter · 2026-08-23
- [source] ChatGPT routing users to 5.5-mini despite selecting GPT-5.6 Sol — devil_ozz · 2026-08-23
- OpenAI routing bug: users report being diverted to GPT-5.5-mini instead of 5.6 — ThutmoseCYH · 2026-08-23
- [source] GPT-5.6 Silently Falls Back to 5.5-Mini for Days Without Acknowledgment — vvebevv · 2026-08-23