Mammouth AI Release Notes

Model Updates ​

• Claude Opus 5.5 replaces Claude Opus 5
• GPT-6 Sol replaces GPT-5.6 Sol
• GPT-6 Luna replaces GPT-5.6 Luna
• Grok 4.7 replaces Grok 4.6

Discover All Models ​

A new button at the end of the model list beneath the prompt box lets you view all available models. Mobile navigation has also been improved.

More Document Formats ​

Mammouth now supports binary file types, including .xlsm, .docm, .jar, and .nes.

Groups for Larger Teams ​

Teams with more than 100 members can now create groups, each with dedicated admins and permissions.

⭐ New Models ​

• GPT-6 Astra
• Claude Fable 5.1

🔄 Model Updates ​

Text:

• GLM 5.3 Flash
• DeepSeek v4.1 Flash

Image:

• GPT Image 2.5
• Grok Imagine Image 2

📂 Document Context Is Now Fully Loaded ​

Previously, when a very long document was uploaded, only part of the document was extracted, which could result in truncated content.

Now, agentic mode is automatically enabled for large document contexts, allowing the full document and its context to be loaded.

🧠 Chat Memory Is Now Larger by Default ​

Several users reported that Chat Memory was too limited, even when they had not manually changed their memory settings. We now automatically enable Large Memory mode for conversations taking place during the current day.

To enable Large Memory mode for all conversations, including those from previous days, you can activate "Extended Memory" mode.

🤖 Smarter Default Model ​

When no model is explicitly selected, the default text model is no longer limited to Mistral model. It now dynamically selects from a set of models based on the prompt.

We have introduced smart model routing for text models.

🔌 Custom Connectors Can Now Be Activated with OAuth ​ …

🔌 Connectors (MCPs) ​

• Pre-built connectors - 4 pre-built connectors (Microsoft 365, GitHub, Notion, and Atlassian) are available directly in the chat.
• Custom MCP - If you don't find the one you are looking for, you can set up your own custom MCP.

Connectors can be activated in the account settings.

For Teams: Team owners can set the connectors for the Team in the permission settings.

🔄 Model Updates ​

• Claude Opus 5 replaces Opus 4.8 and Fable
• Gemini 3.7 Flash replaces 3.6 Flash
• Grok 4.6
• GLM 5.3
• Qwen 3.8

🛠️ Platform Stability ​

We invested a lot in improving the overall platform stability. Many bugs were fixed or largely reduced, including:

• Prompt interruption - Some prompts used to get interrupted before the AI finishes the answer. Not only it's fixed, you can now also refresh the page, turn off the browser or burn your computer, and the streaming will still be continued, unless of course if you click on the stop generation button.
• XLSX upload - Some XLSX files were not uploaded properly.
• Generated files blocking conversations - Some generated files, in particular zip and tar, could block a conversation and prevent new messages from being sent. This is now fixed. …

🎵 Music Generation with Lyria ​

Music Generation - Music generation is now available in Mammouth with Lyria!

🔄 Model Updates ​

• Claude Sonnet 5 replaces Claude Sonnet 4.6
• GLM 5.2 replaces GLM 5.1
• Qwen 3.7 replaces Qwen 3.6
• Grok Imagine Image Quality replaces Grok Imagine Image
• Recraft v4.1 replaces Recraft v4
• Recraft v4.1 SVG replaces Recraft v4 SVG

✨ New Features ​

Agentic Mode - Agentic Mode can be activated in the input. Agentic mode is best for complex tasks such as document generation, document editing and deep documented web searches.

AI Text To Speech - AI Text To Speech is now supported in German and 10 other languages thanks to OpenAI's TTS. French and English TTS are still supported with Voxtral. Technical issues with TTS were also fixed.

🧷 For Developers ​

New models in the API - We added new models in the API. Check now: model-explorer.mammouth.ai

Mammouth Code recommended model - In Mammouth Code, in the model selector, you can now select a model called "mammouth-recommended" and you'll be able to use the same efficient model that we use internally to code. Currently it's GLM 5.2 and we'll update it anytime a new model makes a better deal. Use the mammouth recommended model to generate great code at the best price.