OpenAI burns money on Codex to buy developer loyalty

Two Codex threads today painted a contradictory picture. One noted usage limits resetting unexpectedly (OpenAI has been offering surprise resets to keep developers engaged), while the other reported a quiet reduction in Codex's model context window from 372k to 272k tokens. OpenAI simultaneously gave developers more access and less capability.

The reset strategy is not accidental generosity. Commenters said directly that OpenAI is burning money before its IPO to gain mindshare and that it's bought a lot of developer goodwill this way, but the question is whether it's sustainable. The context reduction is a cost-cutting move dressed up as nothing, which is the kind of thing developers notice and remember.

A separate thread noted Claude Code now ships with a bundled Bun runtime rewritten in Rust, with Anthropic using AI to substantially rewrite the codebase. The tooling war between OpenAI's Codex and Anthropic's Claude Code is intensifying, and the battleground is developer workflow lock-in.