Large Language ModelText GenerationMistral Family

Mistral FT

Mistral FT refers to fine-tuned variants of Mistral AI base language models, exposed via the Mistral API for domain- or task-specific use. These models inherit architecture, context window, and most capabilities from their underlying base models (e.g., Mistral Small, Mistral Medium, Mistral Large) while adapting behavior to customer data. There is no single canonical "Mistral FT" checkpoint with unified public benchmarks; performance depends on the chosen base model and fine-tuning setup.

by Mistral AI
Context Window
32K
API Access
Available
Fine-tuning
Supported

Key Capabilities

  • +Task- and domain-specific fine-tuning on top of Mistral base models
  • +Improved adherence to custom style guides, tools, and workflows
  • +Can leverage same context window and latency characteristics as underlying base model
  • +Supports typical LLM tasks such as chat, summarization, extraction, and code generation
  • +Available via Mistral AI API with managed hosting

Limitations

  • -"Mistral FT" is not a single standardized public model with fixed specs or benchmarks
  • -Quantitative performance varies with the underlying base model and fine-tuning data quality
  • -No consolidated public benchmark suite exists specifically for generic Mistral FT models
  • -Fine-tuning access and configuration details may be limited compared with fully open-source models

Benchmark Performance

reasoning

reasoning

Massive Multitask Language Understanding

81.2%

coding

coding

HumanEval

73.0%

math

math

Grade School Math 8K

81.0%
math

MATH

45.0%

conversation

conversation

Chatbot Arena Elo

1158.0Elo

Alternatives & Comparisons

Open-weight instruction-tuned base model from Mistral with public benchmarks and community tooling.

Strengths
  • + Open weights and permissive license
  • + Good instruction following for its size
Weaknesses
  • - Smaller than frontier proprietary models
  • - Lower performance on complex reasoning than larger models

Sparse MoE model from Mistral with strong performance per dollar and open weights.

Strengths
  • + High quality for cost
  • + Open weights
Weaknesses
  • - More complex deployment due to MoE
  • - Still behind the very largest proprietary models

Fine-tuned variant of OpenAI's flagship GPT-4o model with strong tooling and ecosystem.

Strengths
  • + Frontier-level performance
  • + Rich tooling and integrations
Weaknesses
  • - Fully proprietary and closed weights
  • - Typically higher cost than smaller open models