Mistral Large 3 (v25.12, Base/Instruct)
Mistral Large 3 (v25.12, Base/Instruct) is a multimodal foundation model developed by MistralAI. It is equipped with a massive scale of approximately 6.75 trillion parameters and a wide context window of 256K.
Parameters
6750.0B
Context Window
256K
License
Apache 2.0
Release Date
2025-12-02
API Pricing
API pricing for this model is not yet available
Strengths
- ・Enormous parameter count of 6.75 trillion
- ・Support for long context of 256K
- ・Flexible Apache 2.0 license
Weaknesses
- ・Load due to huge model size
- ・Lack of detailed performance metrics
- ・Possibility of high resource consumption
Use Cases
- ・Advanced analysis of large-scale data
- ・Processing of long documents
- ・Development of multimodal AI
Deep Analysis
Architecture
Granular MoE (675B total, 41B active)
State-of-the-art multimodal
Context Window
256K tokens
API Pricing (Input/Output)
$0.50 / $1.50 per 1M tokens
Release Date
December 2, 2025
License
Open weights
Modalities
Text + Vision
Multimodal
Strengths
- ・Frontier multimodal capabilities
- ・675B total parameters with efficient 41B active
- ・256K context window
- ・Open weights
- ・Strong multilingual support
Weaknesses
- ・Higher pricing than smaller Mistral models
- ・Requires significant compute for self-hosting
- ・Competes with many strong alternatives
Competitor Comparison
| Model | Arena | SWE | GPQA | Price |
|---|---|---|---|---|
| Claude Sonnet 4.6 | - | - | - | Higher |
| GPT-5 | - | - | - | Higher |
| DeepSeek V3 | - | - | - | Lower |
Mistral Large 3 is Mistral AI's flagship open-weight multimodal model with 675B total parameters (41B active). Released December 2025, it features a granular MoE architecture with 256K context and competitive pricing.
Analysis generated: 2026-05-24