Stealth Model Swap? GPT-5.5 High Claims Knowledge Cutoff is June 2024

Hi everyone,

I’m noticing some highly inconsistent behavior across my accounts that strongly suggests behind-the-scenes model routing or dynamic downgrading is happening with GPT-5.5.

My Setup:

• Plus Account (Web UI)

• Business Account (Web UI & Codex workflow)

The Issue: When using the Plus Web UI and Business Codex , I explicitly select the GPT-5.5 High model. However, when I ask a simple system check: “What is your current training data cutoff?”

The model responds with:

“My training data cutoff is June 2024. I can also browse or use tools for current information when needed.”

June 2024 is completely wrong for a brand-new foundational model like GPT-5.5. It feels like the system is stealthily routing my queries to a much older, lower-tier model base despite the UI indicating that “High” is active.

The Discrepancy: To make things weirder, if I ask the exact same question in the Business Web UI , the model responds with:

“My training data cutoff is August 2025.”

The August 2025 cutoff seems to align more with a different variant (potentially the Instant model architecture), but it raises a huge red flag as to why the “High” tier in Codex and Plus Web is giving me a 2024 baseline.

The Impact: The reasoning and output quality have taken a massive nosedive over the past 48 hours. Coding logic is frequently broken, and the model behaves nowhere near the capabilities demonstrated at launch.

Is anyone else experiencing this June 2024 cutoff on their GPT-5.5 High instances? OpenAI team, can we get some official clarification on what is actually running under the hood right now?

Thanks.