Mammouth AI is a French platform that puts text, image, video and web-search models from several AI labs behind one chat interface, with plans sized by usage quota rather than by model access.
Used this tool? Rate it
Used this tool? Rate it
Mammouth AI is a multi-model AI subscription: one web interface where you pick text, image, video and web-search models from different AI labs instead of paying each provider separately. The company describes it as a platform that gives you access to the best AI models on the market in a single interface.
It is run by Mammouth AI SAS, a company incorporated in France with its registered office in Paris. The company says it earns money only from subscriptions and never sells or monetizes user data. A third-party AI tool directory estimated about 2.3M monthly visitors as captured on September 28, 2026. Mammouth works as an AI model aggregator rather than a model maker, so the practical questions are which models are in the catalogue today, how much usage each plan buys before a lighter model takes over, and what happens to prompts once they reach the underlying providers.
As captured on September 28, 2026, every paid plan card lists the same catalogue: text models from Claude, GPT, Llama, Grok, Mistral, Gemini, DeepSeek, Qwen and Kimi, reasoning models, Perplexity for web and deep research, and image and video generation with Nano Banana, GPT, Stable Diffusion, Recraft, FLUX, Sora, Veo and Kling. Plans differ in usage allowance, not in which of these families you can open.
The catalogue changes often. The release notes dated September 18, 2026 list GPT-6 Astra and Claude Fable 5.1 as new models. That same entry states that Sora has been removed following its decommission, while the plan cards captured on the same day still name Sora among the video models. When no model is chosen, the default text model now selects dynamically from a set of models based on the prompt instead of being limited to a Mistral model. Chat memory was also enlarged: Large Memory mode is now on by default for conversations from the current day, and an "Extended Memory" mode extends it to earlier conversations.
Reprompting sends the same prompt to another model with one click, without rewriting it or pasting it into a new conversation, which makes side-by-side comparison the core workflow. One detail matters for longer threads: if you reprompt a later message, the whole conversation history is sent to the new model, not just the last question.
Agentic mode is switched on from the input area and is aimed at document generation, document editing and deep web research with sources. Since the September 2026 release, agentic mode is enabled automatically for large document contexts so that the full document can be loaded instead of a truncated extract.
Four pre-built connectors can be switched on in one click: Microsoft 365, GitHub, Notion and Atlassian (Confluence and Jira). Beyond those, you can register your own MCP server, but it has to be reachable from the public internet. Mammouth only uses the Streamable HTTP transport for these servers, calling initialize and then tools/list.
The developer side is an OpenAI-compatible chat completion API, so existing OpenAI client code can point at Mammouth's endpoint. Each subscription includes a monthly API credit allowance (see Pricing). A shortcut model called mammouth-recommended routes requests to whatever model Mammouth currently rates best for its price, and you pay the same as the underlying model with no markup. Per-token prices in the API table are upper bounds: the actual charge can be lower depending on provider availability, but never higher than shown. Mammouth Code is a terminal-based coding agent; it requires a Mammouth API key and the project is published as open source. It injects system prompts dynamically depending on the task, which the documentation says explains roughly 10k additional tokens per request.
A first session follows the layout of the chat screen:
Consumption is not a simple message count. Longer messages and responses, attached documents and images, tool use such as web search, stronger models and the length of the conversation all raise the cost of each exchange, because accumulated context is sent with every message.
Mammouth's own model guide maps tasks to model categories, which gives a practical picture of where the subscription is used:
The common thread is comparison: the same brief can be run through a second model for a draft, an image style or a fact check without leaving the conversation.
The quota policy sizes each plan against a user profile, and the percentages below are vendor-reported:
For organisations, team accounts use three roles (owner, admin, member), and SSO is available with Microsoft Entra ID, Okta, Keycloak and other SAML 2.0 identity providers. Members can be invited individually or in batches of up to 100 by pasting an email list. Plans can be mixed (Starter, Standard and Expert) within the same team. Inside a team, each user's conversations and documents stay private unless a conversation or a custom Mammouth is explicitly shared. A shared custom Mammouth can be read-only or editable, and sharing it does not expose the conversations held inside it. Owners can also disable specific AI providers, API access or video generation for the whole team.
It is a weaker fit if you need a native iPhone app from the App Store, a guaranteed fixed number of messages with a specific top-tier model, or the option to try a paid plan and get a refund afterwards (see Limitations).
Mammouth AI pricing has three individual plans. The table reflects the pricing page as captured on September 28, 2026, which displayed US dollar amounts excluding VAT.
| Plan | Monthly billing | Annual billing | Usage vs Starter | API credits included |
|---|---|---|---|---|
| Starter | $12/month | $10/month ($120 billed in advance) | Reference | $2 |
| Standard | $24/month | $20/month ($240 billed in advance) | 3× | $4 |
| Expert | $72/month | $60/month ($720 billed in advance) | 10× | $10 |
The pages disagree on currency. The homepage headline advertises access "from €10/month", while the pricing page captured for this review showed dollar prices excluding VAT; neither page states which visitors see which currency. The quota policy adds that the Expert plan offers 10 times more quota for only 6 times the Starter price.
Team plans charge the same price per user as individual plans and add an admin console, centralized billing and shared projects. Enterprise quotes start at more than 100 licences and add smart onboarding, SSO and payment by wire transfer. Online checkout accepts card, PayPal, Klarna and SEPA direct debit (Europe only). Bank transfer requires an annual commitment, and each licence addition then needs a new quote processed within 72 business hours. Plan changes are billed pro rata for the month. The terms describe pre-billing at the start of each period, auto-renewal at then-current rates and no refunds for partial periods.
Against both, Mammouth's distinguishing points are its EU base, its published data-handling commitments for providers and its included API credits; which matters more depends on whether you need a free tier or a wider model list.
On a third-party review platform, Mammouth held a 4.1 score from 265 reviews as captured on September 28, 2026, with 60% five-star and 25% one-star ratings, and the platform notes that the company has not invited customers to review. The platform's AI summary of 59 recent reviews praises the value of one subscription for many models, but mentions slow customer support responses and concerns over whether some models fully match the advertised versions. On the identity point, Mammouth states that models do not know their own identity from training data and that its identity instructions may not apply, for example inside a custom Mammouth.
The refund policy states that no refunds will be issued once you have started using Mammouth, because each interaction consumes API tokens bought from third-party providers, and starting to use the service waives the 14-day EU withdrawal right. Remedies remain available if the platform itself is defective or unavailable, but poor or unexpected AI responses do not count as a service defect. Eligible refund requests go to support from the account email and receive a response within 10 business days. The terms are governed by French law, with the courts of Paris as the jurisdiction.
No free tier appears on the pricing page checked. The terms only say that beta features or free trials may be offered under specific circumstances, without warranty.
To cancel, go to account settings, choose "Manage my subscription" and then "Cancel subscription". Cancelling does not trigger a refund for a period already started.
No. Besides the credits bundled with plans, you can also subscribe on a pay-as-you-go basis directly from the API settings.
Account deletion sits under account settings, then "Advanced settings" and "Delete my account". Deleting a conversation removes its messages and attachments, subject to the 30-day abuse-detection retention described above.
