• Fix thought process for custom OpenAI-compatible providers
• Add "reasoning effort" option in model settings

• Fix Azure OpenAI
• Fix Vercel Gemini image generation
• Add Google Vertex AI support

• Disable reasoning effort for Gemini image models
• Improve support for image models from custom providers like AiHubMix

• Support Gemini 2.5 Flash Image Preview (image generation) from OpenRouter
• Potentially fix Azure OpenAI error

• Linux: partially fix theme switching from app settings
  (e.g change to dark, but still won't react to OS theme change, requires re-opening the app to take effect)
• Add Google Gemini 2.5 Flash Preview (Image Gen)

• New: Linux .deb .rpm (x64) support (no auto-update support)
• New: Reasoning effort option for Groq gpt-oss-20b gpt-oss-120b
• New: Native web search for Groq gpt-oss-20b gpt-oss-120b using Browser Search
• New: New text-to-speech model ElevenLabs eleven-v3
• Fix: Youtube video transcript fetching

• Right-click assistants in the left sidebar to open assistant settings
• Keep main window always running, cmd-w (mac) / ctrl-w (windows) will only hide main window instead of destroying it
• macOS: Add "Auto-hide dock icon" option in app general settings, by default we show both dock icon and menubar icon, you can enable this option to automatically hide dock icon when all windows are closed

• Fix Vercel provider system message error
• Fix high cpu usage when SSE MCP server is used (potentially fixed)

• New provider: Z.ai
• Fix MCP error, #941

• Support MCP OAuth (for sse and http mcp servers, like Replicate MCP and Notion MCP)
• Allow to disable specific tools from an MCP server
• Add experimental Vercel provider (Vercel AI Gateway)

Fix: use Responses API for OpenAI models when reasoning summary is enabled (it's necessary for showing reasoning summary)

Fix minor styling issue.

• Add option for max title lines to display in sidebar
• Fix incorrect API key description in GitHub Models settings

• Fix: only use Responses API for OpenAI models when native search is enabled
• Fix: disable reasoning effort for OpenAI gpt-5-chat
• Fix: disable OpenAI reasoning summary when reasoning effort is turned off

• Support reasoning effort in OpenAI GPT-5
• Fix max output tokens with OpenAI GPT-5
• Support reasoning summary for OpenAI GPT-5
• Enable native web search for OpenAI models

Fix OpenAI GPT-5 temperature setting.

Fix: show Ollama GPT-OSS thinking content

Support chatwise://mcp/add?json=<base64 encoded mcp json config> for adding MCP servers

• New provider: Poe (https://x.com/poe_platform/status/1950967460125131047)
• Allow to hide menubar icon on macOS:

• Support fetching ZhiPu (bigmodel.cn) models, open sourced at https://github.com/egoist/chatwise-models
• Fix Anthropic oauth status display error
• Fix assistant sorting on Windows

Fix Claude Pro/Max error.

Claude Pro/Max

Now you can authenticate Anthropic with your Claude Pro/Max subscription

Tray/Menubar

Now ChatWise lives in your system tray or menubar, click the icon to open window, or right click for more options

More

• You can use Command+P (mac) or Ctrl+P (windows) to open the command center now, the old hotkey Command+K still works
• Fix spinner bug

• New: allow to specify model in prompt settings
• Fix: JSON comment support when importing mcp from clipboard

Fix settings window glitching on macOS.

• OpenRouter: allow adding custom models
• Together AI: allow adding custom models
• New provider: Hyperbolic
• New Provider: Cerebras

• fix web search: change default openrouter web search model to openrouter-qwen/qwen3-14b:free
• fix image support: support uppercase filename

Auxiliary Visual Model

For models that don't support image input, like Kimi K2, you can select an auxiliary visual model in Settings > Extensions > Auxiliary Visual Model to get visual description for the images you attached

Docs: https://docs.chatwise.app/auxiliary-visual.html

Fix typo.

• Fix: remove instruction when using custom openai model in text-to-speech settings
• New: add groq to text-to-speech settings
• New shortcuts: CommandOrControl+Shift+C to copy recent response, CommandOrControl+R to regenerate response. (Command on Mac and Control on Windows)
• Add an option to hide reasoning content by default

• tts: allow to use a custom model for the openai provider
• chatwise search: previously when a web page is too long it would use the first 4000 characters, now we added an option to summarize web pages with AI first
• new provider: 302.ai