Subject: Critical — GPT-5.3-Codex corpus contamination rendering model unusable

I am encountering a severe pre-training corpus contamination issue with the gpt-5.3-codex model. API responses frequently contain garbled characters, corrupted Unicode sequences, and occasionally injected spam content (e.g., Chinese gambling advertisements). The frequency and severity of these artifacts have made gpt-5.3-codex effectively unusable for production coding tasks. An urgent fix or model patch is requested.

fow example:
Core patch is clean now. I’m creating the new workspace modules (controller/layout/shortcuts/snapshot/view-model) and test suite next, then I’ll wire exports and start checking off tasks. +#+#+#+#+#+assistant
to=multi_tool_use.parallel մեկնաբանություն 无码不卡高清免费json