ChatGPTBox-dev/chatGPTBox v2.7.0

on GitHub

Here is another feature and maintenance release for ChatGPTBox.

This release adds four built-in API providers, refreshes major model presets, makes temperature overrides opt-in, and delivers a broad set of streaming, configuration, and browser-integration reliability improvements.

Existing API mode selections, custom provider settings, and API keys are migrated automatically.

Compatibility note: ChatGPT (Web) is no longer enabled by default, because persistent login and unusual activity failures have made the reverse-engineered integration increasingly difficult to maintain reliably. Existing selections are preserved. The obsolete Poe Web integration has been removed, and existing Poe Web modes and sessions are no longer available.

Changes

Features

• Add Google Gemini, xAI, NVIDIA NIM, and Mistral AI as built-in OpenAI-compatible API providers, with dedicated API key setup and request routing.
• Add an explicit Override provider temperature setting. By default, temperature is omitted so each provider or model can use its own default, while known incompatible models never receive the parameter.

New Models

• Add OpenAI GPT-5.6, including the GPT-5.6 Sol, Terra, and Luna tiers.
• Add Anthropic Claude Sonnet 5 and Claude Opus 5, with Claude Opus 5 as the default Opus-tier API option.
• Add xAI Grok 4.3, Grok 4.5, and Grok 4.6.
• Add Google Gemini 2.5 and 3.x model presets, with Gemini 3.1 Pro Preview and Gemini 3.7 Flash enabled by default.
• Add Mistral Medium / Small / Large and selected NVIDIA NIM models, including Nemotron 3 Super and Nemotron 3 Ultra.

Improvements

• Refresh the default-enabled API modes with current OpenAI, Anthropic, Google, xAI, Mistral AI, NVIDIA NIM, and OpenRouter choices. OpenRouter Auto Router and Free Models Router are now enabled by default.
• Preserve configured API mode selections when defaults change, and allow future defaults to be added to customized mode lists without restoring choices that users previously removed.
• Allow custom provider endpoints to be edited, improve the provider editor, and preserve provider identity, API keys, and historical conversation routing when provider IDs are migrated or become reserved.
• Improve request error reporting by distinguishing invalid endpoints, request transport failures, and interrupted response streams, with localized and more actionable diagnostics.
• Improve Claude streaming and completion handling so refusals, response limits, token limits, and incomplete streams are reported correctly instead of appearing to finish successfully.
• Add ChatGPTBox app attribution headers to requests sent directly to OpenRouter, without forwarding them to third-party proxy endpoints.
• Improve input resizing on the independent conversation page with pointer and keyboard controls.

Fixes

• Fix conversations remaining stuck on Stop when requests are slow, cancelled, or interrupted by a proxy-tab disconnect, while preserving partial streamed answers where possible (#995).
• Fix server-sent events being lost when CRLF delimiters are split across response chunks (#1036).
• Fix Enter used to confirm an IME candidate accidentally sending a message or stopping the current response (#1038).
• Fix rapid storage updates overwriting earlier configuration changes (#1037).
• Fix asynchronous edge cases involving background fetches, extension tab opening, Close All Chats In This Page, and deferred input focus (#1004, #1014, #1015, #1018).
• Fix Brave and DuckDuckGo integrations giving up before dynamically rendered search-result containers appear (#1016).
• Fix GitHub issue and pull request URL variants being missed or unnecessarily remounting summaries, and reject malformed nested thread paths (#1039, #1046).
• Fix fetch-based GitHub and GitLab analysis treating HTTP 4xx or 5xx error pages as valid content (#1035).

Removed

• Remove the obsolete reverse-engineered Poe Web integration, including its models, sign-in flow, streaming, and session handling.

Security

• Replace jsonwebtoken and its browser crypto polyfill chain with focused HS256 signing backed by @noble/hashes, reducing bundled crypto dependencies while preserving existing token behavior.
• Harden the Firefox metadata workflow against command injection by validating and safely passing manually supplied version values.
• Apply non-breaking npm audit updates for multiple transitive dependency advisories.

Chores / Developer Experience

• Standardize Safari builds on the canonical ChatGPTBox name, make packaging reproducible from the committed lockfile, and add app-signature and DMG integrity verification to pull request CI.
• Add Firefox metadata recovery and update workflows, including retry handling while newly submitted AMO versions become available.
• Improve search-engine configuration verification reliability and execution limits without changing extension runtime behavior.
• Expand automated test coverage across configuration migration, provider routing, streaming, browser operations, and UI behavior.
• Refresh project documentation, privacy wording, translated READMEs, localization guidance, GitHub Actions dependencies, and maintenance automation.

Contributors

A huge thank you to everyone who contributed to this release through code, bug reports, reviews, testing, and ideas.

Full Changelog: v2.6.1...v2.7.0