Mistral Large 4
Mistral’s flagship for coding, agents and document analysis, now in public preview.
At a glance
- Architecture
- 1.05T parameters
- 49B active per token.
- Context window
- 524,288 tokens
- On MiniRouter’s Vercel route.
- Input and output
- Text → text
- Tools and reasoning supported.
- Maximum output
- 262,144 tokens
- Within the shared context window.
Publisher pricing
Preview rates per million tokens, before MiniRouter markup.
- Input
- $0.68
- Output
- $2.09
Cached input is $0.07/M. Preview rates are 50% off Mistral’s standard $1.36/M input, $0.14/M cached input and $4.18/M output, with no end date stated.
Reported benchmarks
From Mistral’s announcement. MiniRouter has not reproduced these results.
- DeepSWE v1.1
- 61.7%
- Terminal-Bench 4.0
- 28.3%
- AutomationBench
- 59.9%
Preview results may change. Mistral has not published full effort and harness settings; comparisons across publishers are approximate.
Sources
- Read at Mistral AI ↗
Mistral Large 4 public preview announcement
Published Oct 6, 2026
- Read at Mistral AI ↗
Mistral Large 4 model documentation
Published Oct 6, 2026