DeepSeek V4 Flash Free Capped at 200K Context in OpenCode Metadata Bug Report

abhisheksharma611 · ghdev · 2026-08-07

A developer filed a bug report on the OpenCode Zen platform revealing that the DeepSeek V4 Flash Free model is incorrectly capped at a 200K context window in the models.dev metadata, despite natively supporting 1M tokens.

This configuration issue forces new users into a lower default limit and triggers premature context compression. While users can manually override this by editing their opencode.json file, the developer requests a server-side metadata update to align the free tier's default with the model's official capabilities.

Original post →

More from coding & agent

coding & agent channel →