TypingMind Release Notes

We have just added Claude Opus 4.8 on TypingMind!

✨ Also in this update:

• Added Claude Opus 4.8 and Opus 4.8 (Fast) via OpenRouter.
• Updated model provider pricing/metadata for several catalogs, including DeepSeek, Deepinfra, Groq, Mistral, and OpenRouter models.

✨ New

• Added new Deepinfra models and OpenRouter models.
• Deprecated/deleted models are now kept as deprecated so old chats can show model status more clearly.

Improvements

• Updated OpenRouter and Deepinfra model metadata to reflect latest limits, pricing, and supported capabilities, including context/output limits, pricing, PDF support, and prompt caching support.
• Updated OpenRouter Anthropic latest aliases to use Claude-specific settings for better reasoning effort and prompt caching support.
• Expanded OpenRouter reasoning effort values: none, low, medium, high, xhigh.
• Removed deprecated DALL-E 3 builtin plugin as this image model is deprecated by OpenAI. Please switch to GPT Image Editor for image generation.

Fixed

• Fixed Fireworks provider issue where previous user input or stale/duplicated assistant content could be echoed when using the OpenAI Responses API format.

Gemini 3.5 Flash, Google’s latest Flash-tier model, is now available on TypingMind!

This models is optimized for coding proficiency and parallel agentic execution loops.

✨ What’s New

Support all provider settings for profiles

• You can now set up API keys for different profiles from all available providers
• Older profile API key assignments are automatically migrated to the new provider settings format.

⚙️ Bug fixes and improvements

• In multi-model chats, clicking a model header/tab now scrolls directly to that model’s latest assistant message. This makes it easier to continue reading the newest response without scrolling through earlier replies.
• Fixed an issue where switching model tabs on mobile could fail to scroll to the latest message if the response card had not mounted yet.
• Removed unsupported metadata fields from tool calls before sending requests to OpenAI-compatible providers.
• Fixed provider API errors caused by extra tool call fields such as output_item_id, call_id, and annotations.

Here's what's new in this update:

✨ New

Redesigned Admin Analytics

• The analytics dashboard now has two tabs: Overview for high-level insights and Exploration for deeper analysis.
• New metrics have been added, including Characters, plus breakdowns by users, models, and agents, with improved charts, styling, and tooltips.
• CSV export for analytics data has been centralized for a more consistent experience.
• Date ranges are now inclusive, normalized to UTC day boundaries, and consistently limited to 1 year.

Exploration mode now supports filtering by metric, including user-level filtering where permitted.

🛠 Improvements

** Web App Builder & Interactive Canvas:**

• AI can now update existing HTML using small patch-style edits instead of regenerating the full document, making follow-up edits faster and more token-efficient.

Artifacts side panel — version labels:

• When a plugin has multiple tool functions, artifact versions are now correctly grouped into one continuous stack and no longer display "undefined - Version N".

⚙️ Bug Fixes

• Fixed DeepSeek API handling so reasoning_content is preserved for supported models.
• Fixed an issue where MCPs/tools did not work with some providers, including Mistral.
• Improved cross-provider tool/function calling compatibility by removing unsupported annotations from function specs before sending tool definitions.
• Improved the Organization dashboard Recent Activity feed with clearer audit log formatting, names, and layout.

We're excited to introduce the TypingMind Referral Program. Starting today, every time you refer a friend who buys a TypingMind license, you both unlock 0.5 GB of free cloud storage — forever.

Here's how it works

• Go to Settings → Cloud Sync & Backup → Refer friends
• Copy your personal referral link (typingmind.com/r/). For example:

https://www.typingmind.com/r/S4G6TB5G

• Share it with friends via Email, X, or LinkedIn — or anywhere else

When they purchase a license through your link, you both get 0.5 GB added to your accounts instantly.

Notes:

• Available for individual TypingMind accounts (typingmind.com)
• No expiry on earned storage — keep it for life
• No referral limit — earn as much as you want
• Your friend must use your link before purchasing for both of you to receive the reward

We’ve expanded official model support beyond OpenAI, Anthropic, and Google.

You can now easily add models from:

• OpenRouter
• DeepInfra
• DeepSeek
• Groq
• Mistral
• Moonshot
• Perplexity
• xAI
• Z.ai

More providers and models will be added soon.

We also moved API keys setting from Settings → API Key to Models → API Keys.

To set up API keys, go to Models from the left-side panel, then switch to the API Keys tab and enter your API key.

You can also go to Models and click the gear icon next to each AI provider to update its API key.

Also in this update:

• Added support for custom endpoints for OpenAI Responses API requests.

What’s new

• Added token limit: Admins can now set up token limit per time period for specific models, AI Agents, users/user groups.
  Go to Admin panel → Under the Access & Limits section, select Usage Limit.

• Added admin toggle for ElevenLabs: Admins can now show or hide the ElevenLabs integration directly from the admin panel settings.

🛠 Improvements & Bug Fixes

• Fixed a flashing issue during the close transition by freezing content to prevent flicker.
• Fixed an overflow issue in attachment and file layouts. Flex layout constraints (min-width + flex-shrink) have been added to keep attachments properly contained.
• Fixed an issue where inline agents could incorrectly search the wrong Knowledge Base tags.
• Fixed artifact dimensions so artifacts now display with the correct height in the UI.

GPT-5.5 and GPT-5.5 Pro are now accessible via TypingMind!

Better at coding, tool use, research and analysis!

Here's what's new in this update:

• GPT Image 2 model support: The GPT Image Editor plugin now supports the gpt-image-2 model, giving you access to OpenAI's latest image generation capabilities directly within the plugin.

• Resizable sidebar: You can now drag to resize the sidebar to your preferred width. The size is saved and persists across app reloads.

• Custom body parameters in model settings: You can now add custom request body parameters directly in a model's settings. This enables use cases such as setting service_tier: "priority" to help reduce 429 rate-limit errors on models like Gemini 3.1 Pro.

Claude Opus 4.7 is now available on TypingMind!

• Better at complex coding work
• More reliable on long tasks
• Check its own work before answering
• Produce more polished professional outputs like UIs, slides, and documents.

Try now!

Other updates:

• Added support for xhigh thinking effort for Claude Opus 4.7
• Treated xhigh as max thinking effort for other Claude models