Mistral Vibe: Default Model Gets New Thinking Levels and a 256k Context Window
Mistral VibeMistral Vibe 2.26.1 changed the default model's thinking levels to off and high, declared a 256k context window with auto-compaction derived from it, and made the off level actually disable reasoning. The bundled local Devstral model was removed, and stale pins to it now fall back to the hosted default.
Key Takeaways
- The default model now offers only off and high thinking levels, simplifying the choice.
- Stored low picks reset to off and stored medium picks reset to high, so existing configs migrate automatically.
- The off level now truly disables reasoning where the model supports it, rather than leaving the server default in control.
- A 256k context window is declared for the default model, and auto-compaction now derives from it.
- The bundled local Devstral model was removed from Mistral Vibe.
- Stale pins to the removed model warn and fall back to the hosted default instead of failing.
Thinking Levels Reworked
Mistral Vibe 2.26.1 changes how reasoning is controlled for the default model. Its thinking levels are now off and high. Existing saved choices migrate automatically: a stored low pick resets to off, and a stored medium pick resets to high.
Off Now Means Off
The off thinking level now actually disables reasoning on models whose offered levels include it. Previously, selecting off could silently leave the server's default in charge, so reasoning could still happen. This fix makes the setting trustworthy for developers who turn thinking off to save time or cost.
Larger Declared Context
The default model now declares a 256k context window, and auto-compaction is derived from that figure. Long sessions can therefore run closer to the model's real capacity before the conversation is compacted.
Local Devstral Removed
The bundled Devstral (local) model was removed. A stale pin to it now shows a warning and falls back to the hosted default model, so existing configurations keep working without manual edits.