← All modelsbalanced
Mistral: Mistral Medium 3.5
Mistral
Capabilities
VisionToolsStructured OutputExtended ThinkingLong ContextCode
Strengths
- Strong long-context support (262K tokens) for document summarisation
- Excellent multimodal vision and image understanding
- Instruction-following and agentic workflow capabilities
- Cost-effective balanced pricing mid-range option
Weaknesses
- Very slow inference speed (0.11 speed_score, ~164 tok/s)
- Weak agentic reasoning and general intelligence benchmarks
- Limited code generation performance relative to specialist models
- Closed-source weights reduce customisation and privacy control
Pricing
Input / 1M tokens
$1.5
Output / 1M tokens
$7.5
Context window
262.144k
Transparency
Open weights
0.0 / 10
Open training data
0.0 / 10
Open methodology
0.0 / 10
Licence openness
4.0 / 10
Provider disclosure
5.0 / 10
FMTI company score
N/A
Composite:5.5 / 10
Mistral publishes model reports but does not open-source weights for Medium 3.5. Licensing terms are proprietary. Provider offers moderate transparency on capability benchmarks but limited disclosure on training data composition or infrastructure.
Sustainability
Inference energy
N/A
Training footprint
N/A
Provider infrastructure
5.5 / 10
Composite:5.0 / 10
No public commitments to renewable energy or carbon offsets. 128B parameter model incurs moderate inference costs. Provider sustainability practices are not widely documented.
Task Fitness
question answering
52%
code
60%
mathematics
85%
business communication
84%
vision
80%
analysis
61%
extraction
81%
long-form generation
78%
research
78%
reasoning
52%
summarisation
82%
translation
75%
conversation
84%