Mistral Vibe: Mistral Large 4 Becomes the Default Model
Mistral Vibe now uses Mistral Large 4 as its default model, replacing Mistral Medium 3.5. Users who prefer the previous model can pin it with active_model = "mistral-medium-3.5", and mistral-large-latest is now treated as Large 4 with a 1,048,576-token window.
Key Takeaways
- Mistral Large 4 is now the default model in Mistral Vibe.
- Users can stay on Medium 3.5 with
active_model = "mistral-medium-3.5". - The
mistral-large-latestalias resolves to Large 4 with a 1M-token window. - Thinking supports low, medium and high, and
offuses low. - Choosing a default thinking level no longer pins it in config.
vibe -p --thinking LEVELsets thinking for one run only.
A new default
With 3.1.3, Mistral Vibe switched its default model to Mistral Large 4. Anyone who has not chosen a model explicitly will now get Large 4, which can change answer quality, speed and cost.
Staying on the previous model
Users who want to remain on Mistral Medium 3.5, or whose own Mistral server does not offer Large 4, can set active_model = "mistral-medium-3.5" in their configuration.
How Large 4 is handled
Vibe treats mistral-large-latest as the model the Mistral API serves under that alias, today Mistral Large 4. That means a 1,048,576-token context window, low, medium and high thinking levels (with off mapping to low), and earlier reasoning preserved in later requests.
Related thinking-level change
Choosing a model's default thinking level no longer pins it in the config, so the model follows later changes to that default. vibe -p --thinking LEVEL also sets thinking for a single headless run without saving it.