Mistral Large 4 entered public preview on 6 October, with hosted access through Mistral’s API and OpenRouter. A Hugging Face community guide says the model has a published one-million-token context window and lists preview API prices, giving developers something concrete to assess beyond the launch-day numbers.
Mistral AI Watch analysis
What happened
The guide says developers can try the model in Mistral Studio or call it through the API using the model ID mistral-large-4; it also gives the OpenRouter ID mistralai/mistral-large-4-0. It lists preview rates of $0.68 per million input tokens, $0.07 per million cached input tokens and $2.09 per million output tokens. The author says these details were checked on 6 October and advises readers to confirm current terms before deploying.
The model is described as a multimodal mixture-of-experts system with 1.05 trillion total parameters and 49 billion active per request, plus a 1.6-billion-parameter vision encoder. The guide distinguishes public-preview API access from downloadable weights: it says the weights are a separate release and are not required for hosted use, while self-hosting depends on weights being published. Read the Hugging Face community guide.
Why it matters
For developers, the immediate choice is between Mistral’s direct API and a third-party route, not between hosted access and a ready-to-download model. The stated context window and token prices help teams estimate whether the preview fits a workload, though a large context limit is not a promise that every detail in a huge prompt will be retrieved correctly.
Our read
This is useful launch-day guidance, not a benchmark or an independent test of model quality. The practical takeaway is to pin the preview model ID, check the current price and access terms, and test it against a fixed set of real tasks before putting it to work. One million tokens is a roomy desk; it does not guarantee the model will find the right paper on it.
What to watch
- Whether Mistral changes preview pricing, access limits or model behaviour.
- When downloadable weights become available, and under what licence.
- How the model performs on independent evaluations and real developer workloads.
Discussion spark: Would you choose a hosted preview with a million-token context window, or wait for downloadable weights and more independent testing before building around it?
Sources and evidence
- How to Use Mistral Large 4: API, OpenRouter, and Setup (6 October 2026, 14:19 UTC)
not affiliated with or endorsed by Mistral AI