OpenAI retires 15 model IDs July 23 — every Codex snapshot included
A hard API cutoff lands Thursday: deep-research jumps to $30/$180, DeepSeek aliases 404 on July 24, and MCP's breaking spec follows July 28.
Copy markdown
15 OpenAI IDs go dark Thursday
OpenAI's deprecation wave hits a hard cutoff on July 23: calls to gpt-5-codex, computer-use-preview, o3/o4-mini-deep-research and 11 other IDs will fail outright, with no grace period. Audit every pinned model string in your configs, CI, and SDK calls now — the replacements aren't drop-in.
Every Codex snapshot retires at once
gpt-5-codex, gpt-5.1-codex, -codex-max, -codex-mini and gpt-5.2-codex all sunset the same day. The full-size Codex models map to gpt-5.5; codex-mini maps to gpt-5.4-mini. Re-run your agent evals before cutover — long-horizon behavior differs from the snapshots you tuned against.
Deep-research gets a price hike
o3-deep-research and o4-mini-deep-research both fold into gpt-5.5-pro at $30 in / $180 out per 1M tokens — a real cost bump for automated research pipelines. computer-use-preview moves to gpt-5.4-mini. Re-benchmark and re-budget before you flip the switch.
DeepSeek aliases 404 on July 24
The day after OpenAI's cutoff, deepseek-chat and deepseek-reasoner stop resolving. Repin to deepseek-v4-flash and deepseek-v4-pro respectively — it's a rename with no capability change, but any unpinned integration starts returning 404s.
MCP's breaking spec lands July 28
The 2026-07-28 Model Context Protocol spec drops the initialize handshake and session management for a stateless core, and deprecates Roots, Sampling and Logging. Beta SDKs — Python v2, TypeScript v2 with a codemod, plus Go and C# — are out now; test against real traffic this week, not just the happy path.