V6 brings ChatGPT Web models into Codex's model and Effort selectors, optional Pro usage estimates, Bigger Context, and control over browser conversations. Version 6.1.15 supports the newly observed model picker, restores Temporary Chat by default, and fixes worktree access and launcher warnings.
Changes in 6.1.15
- Updated ChatGPT model picker. Selects the exact GPT-6 Sol or Astra row and supports the new six-position controls. The requested model and effort are verified before sending. Earlier supported picker layouts remain supported (#921, #923).
- Temporary Chat returns, enabled by default. Turn off Settings → Temporary Chat to use ordinary saved conversations. An explicit opt-out is preserved. If a connector is unavailable in Temporary Chat, the error points to this setting (#918).
- Think works with the renamed icon. Recognizes ChatGPT's current Think control across languages without depending on the English button label (#902).
- Full Access works in managed worktrees. An authenticated native Full Access task can use its current worktree even when it is outside the original retained workspace roots. Restricted permission modes continue to enforce their verified roots.
- Session warnings stay above the browser. The sign-in reminder is visible above ChatGPT without moving it to a footer or reducing the browser's size.
- Doctor explains conflicting compaction settings. Warns when global or default-profile context and compaction overrides exceed the selected Web model's limits, while leaving the settings unchanged (#919).
- Safer startup and settings recovery. Stops a newly started owned runtime if recording ownership fails, and preserves launcher settings when their JSON file contains a UTF-8 byte-order mark.
- More reliable turn identification and diagnostics. Rendered line breaks no longer invalidate a correctly submitted turn marker. Diagnostics record stream completion and observed UI structures without message text or tool arguments. UI contracts now have separate owners, dated evidence and regression fixtures.
- Clearer documentation. Explains why one oversized message can exceed a Bigger Context stage even when the total window is larger, and removes the blanket promise that Web requests never affect Work or Codex usage (#874, #922).
V6 features and earlier fixes
- Models and effort. Account-supported GPT-5.6 and GPT-6 models appear in Codex's native picker. Sol reasoning uses Effort, Instant keeps a separate context budget, and Free accounts expose Luna/Think. Earlier patches improved translated controls, Pro verification, locked options and Browser Smoke cancellation.
- Bigger Context and long tasks. Large histories use one, two or six ordered messages. GPT-6 Sol supports configured windows up to 120,000 tokens on Plus and 240,000 on Pro at supported reasoning levels; Instant keeps standard context. These are bridge context budgets, not guaranteed upstream message limits. Earlier fixes preserve the accepted task, active goal, date and delegation context through compaction, handle formatted user messages, and prevent repeated uploads of the same rejected transaction. Luna/Think Bigger Context remains experimental.
- Chat controls and Pro estimates. Optional fresh chats rebuild the same Codex task in a new browser conversation. Optional next-message suggestions use GPT-6 Sol Medium when the context fits. The Limits tab estimates the launcher's own submissions for supported Pro plans; it does not report OpenAI's remaining allowance or messages sent elsewhere.
- Tools and replies. Full harness mode provides native files, commands, tools and approvals through MCP. Command polls are bounded to 60 seconds per MCP call. Earlier updates improved permission guidance, tool settlement, cancellation, Markdown fidelity and delayed-final handling without resubmitting accepted work.
- Diagnostics and desktop reliability. Failed browser captures are retained for up to three days within a 100 MB budget, with active captures protected. Earlier patches improved passkey sign-in, account switching, system proxy support, Windows certificate handling, update rollback and downgrade prevention. The launcher uses Electron 42.11.16 and pinned Bun 1.4.0, with Windows x64, macOS Apple silicon/Intel and Linux x64/ARM64 packages.
Windows tunnel installation in #870 and the context-overflow report in #913 still need the requested reporter evidence. The separate long-running recovery work in draft PR #914 is not included. ChatGPT account limits and safety refusals still apply; shared safety reports remain tracked in #622.
Thanks to @nofihq for the model-picker fix and earlier compaction, delegation, date-handling and dependency work; @lce52147 for Think and Temporary Chat evidence; @thomastsaidotcom for earlier smoke-test reproductions; and everyone who supplied logs and tested fixes, including @iminfine, @zmahomedy, @alexzwy51 and @martjay.
The Pro Researcher skill is available for delegating research to Pro sub-agents.
Update
Finish active tasks, quit the launcher, install the update and reopen it. Restart Codex after updating its integration. Existing launcher settings and the ChatGPT profile are preserved; check Settings → Temporary Chat if you prefer ordinary saved conversations.
If a closed issue still happens, open a new issue with Activity → Export safe log, the failure time, launcher/Codex versions and model/effort. If your issue remains open, add the new results there.