← Back to directory
Mistral AI
Mistral Medium 3
Released May 7, 2025 · knowledge cutoff 2025-01
- Status
- Superseded
- Location
- France
- Modality
- Multimodal
- Context window
- 131K
- Max output
- 8K
- Speed
- —
- Price ($/MTok in / out)
- $0.40 / $2
- Cost per task
- —
- Open weights
- No
Benchmarks3 of 8 reported
MMLU-Pro
77.2%
GPQA Diamond
57.7%
LMArena Elo
1330
SWE-bench Verified
—
Terminal-Bench 2.1
—
AIME (latest)
—
Humanity's Last Exam
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Deploys inside your own VPC or on-premises, which is the entire point for regulated buyers
- Consistent, unfussy output for summarising, drafting and translation
- Low latency for its capability class
Weaknesses
- Closed weights broke the expectation Mistral's open releases had set
- No deliberation mode, so hard reasoning is simply out of scope
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor