Few interesting changes in 18.3.0:
This is a dev post classified by Jev as AI dev tools (a launch), kept by the Dev Radar because it carries real work, not commentary.
Few interesting changes in 18.3.0: - Reduced base token usage from ~30k to 10k (!!) You should have a much more pleasant experience with local models wrt: prefill. - Multiple tools were folded into read/write/bash eliminating tool overhead. - Skill descriptions are now compressed by default. - Added detection for Codex cyber access entitlements, explicitly requesting it even for non-daybreak models, which should get you less refusals if eligable. - Added support for Apple Foundational Models. (8K context so only viable as a tiny model for the time being) - Offline token counting for TypeS
Posted by Can Bölük (18k followers) 1 h ago · 123 likes · 2.6k views · view the original post on X. Kept by the Dev Radar as AI dev tools.
More dev work like this
- 如果你经常用 AI Agent 写代码，又不太想把代码和文件交给云端，可以看看 Magnitude。 — @bkdgiffug
- Switching coding agents shouldn’t mean re-explaining the whole job. — @DanKornas
- PStack go brrrrrr. I think I've unlocked a monster with Cursor Projects. All while… — @RayFernando1337
- There are some new top models leading our code review benchmark. — @Macroscope
- Day 1 of @WeAreDevs is a wrap! — @pamelafox
- Your agent was asked to export a file. It also wrote a checksum nobody uses. — @DanKornas
- Nothing in the world could have prepared me for the root cause of the performance issues… — @theo
- Pipelex executes typed AI procedures declared in .mthds files. — @tom_doerr
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 23.1k posts from 5k X accounts over the last 21 days, 2.7k tools, 12 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-24 02:15 UTC. Full method.