github miuuyy/codex-chatgpt-web v6.1.19

46 minutes ago

Changes in 6.1.19

  • Bigger Context keeps requests in one message when they fit. The launcher checks the complete request against the selected model’s limits before splitting it. Larger requests can use two or six messages. When reconstructing a chat for compaction, the bridge tries to keep all history before trimming older records. Thanks to @kstroevsky and @mvclaudianobj for documenting the affected cases.
  • The model picker explains usage limits. When ChatGPT disables reasoning options, Setup, the smoke test and task turns report the limit. An explicitly selected GPT-6 Instant is verified again before sending. A slow-loading slider can still appear normally. Thanks to @KoscheiiB for the picker capture in #993.
  • Images and restored drafts. The launcher recognizes image thumbnails that have no preview button and stops repeating the same failed upload. A saved composer draft is replaced before staging a new prompt (#996, #977).
  • Long tasks and replies. Compaction now recognizes the native list of subagent IDs in a midnight environment update. Streaming Markdown replies can finish when ChatGPT redraws a Mermaid preview (#982, #974).
  • Sign-in and browser settings. macOS passkey sign-in can import larger browser storage values (#991). The embedded browser follows the selected language after restart and has a separate option to reset its session data. Safe logs distinguish a signed-out ChatGPT session from a malformed expiry. Thanks to @Mjstq for the browser settings changes.
  • Setup and dependencies. Setup opens the reasoning menu on the affected picker layout before inspecting the slider. This release also updates the Hono dependencies.

Earlier changes in V6

V6 integrates account-supported Web models and effort levels into the native Codex picker. The launcher includes optional Pro usage estimates, Temporary Chat settings, fresh chats per turn, composer suggestions and Linux ARM64 packages. Earlier 6.1 patches improved worktree and goal continuity through compaction, staged uploads, translated model controls, passkey sign-in, updater diagnostics and recovery from interrupted turns.

Install or update

Finish active tasks, quit the launcher, install the update and reopen it. If paid Web models are missing, use Setup → Install into Codex → Reinstall and fully restart Codex.

If a fixed issue recurs, open a new issue with Activity → Export safe log, the failure time, your launcher and Codex versions, and the model and effort you used.

Don't miss a new codex-chatgpt-web release

NewReleases is sending notifications on new releases.