Skip to content

Devstral 2 123B

by Mistral AI · mistral family

123B

parameters

text-generation code-generation reasoning math

Devstral 2 123B is Mistral AI's flagship code-focused model, designed for advanced software engineering tasks including code generation, debugging, and complex reasoning about codebases. With 123B parameters and a massive 262K context window, it can process entire repositories at once. This model represents a significant step up in coding capability from its predecessors, excelling at multi-file code understanding, architectural reasoning, and generating production-quality code across dozens of programming languages.

Quick Start with Ollama

ollama run q4_K_M
Resources Ollama Hugging Face Official Page
Creator Mistral AI
Parameters 123B
Architecture transformer-decoder
Context 256K tokens
Released Dec 10, 2025
License Mistral Research License
Ollama devstral-2

Quantization Options

Format File Size VRAM Required Quality Ollama Tag
Q4_K_M rec 62 GB 67 GB q4_K_M
Q8_0 123 GB 129 GB q8_0

Compatible Hardware

Q4_K_M requires 67 GB VRAM

Compatible Hardware

HardwareVRAMTypeFitEst. Speed
Mac Studio M4 Ultra 512GB512 GBmacRuns~12 tok/s
Mac Pro M2 Ultra 192GB192 GBmacRuns~12 tok/s
Mac Studio M4 Ultra 192GB192 GBmacRuns~12 tok/s
Mac Studio M4 Max 128GB128 GBmacRuns~8 tok/s
MacBook Pro M4 Max 128GB128 GBmacRuns~8 tok/s
MacBook Pro M5 Max 128GB128 GBmacRuns~8 tok/s
NVIDIA RTX PRO 6000 Blackwell96 GBgpuRuns~29 tok/s
MacBook Pro M3 Max 96GB96 GBmacRuns~6 tok/s
Mac mini M4 Pro 64GB64 GBmacCPU Offload~1 tok/s
Mac Studio M4 Max 64GB64 GBmacCPU Offload~2 tok/s
MacBook Pro M4 Max 64GB64 GBmacCPU Offload~2 tok/s
MacBook Pro M5 Max 64GB64 GBmacCPU Offload~2 tok/s
NVIDIA RTX 6000 Ada Generation48 GBgpuCPU Offload~4 tok/s
NVIDIA RTX A600048 GBgpuCPU Offload~3 tok/s
NVIDIA RTX PRO 5000 Blackwell48 GBgpuCPU Offload~4 tok/s
Mac mini M4 Pro 48GB48 GBmacCPU Offload~1 tok/s
MacBook Pro M3 Max 48GB48 GBmacCPU Offload~2 tok/s
MacBook Pro M4 Pro 48GB48 GBmacCPU Offload~1 tok/s
MacBook Pro M4 Max 48GB48 GBmacCPU Offload~2 tok/s
MacBook Pro M5 Max 48GB48 GBmacCPU Offload~2 tok/s
MacBook Pro M5 Pro 48GB48 GBmacCPU Offload~1 tok/s
86 hardware device(s) cannot run this model at Q4_K_M.

Benchmark Scores

81.0
mmlu