Devstral 2 123B Instruct
Parameters
-
Context Length
262K
Modality
Text
Architecture
Dense
KV Cache Rate
352 KB/tok
License
Modified MIT
Release Date
1 Dec 2025
Knowledge Cutoff
-
API Pricing (per 1M)
Input: $0.00 · Output: $0.00
Hardware Sizing
Precision
Context
Workload Memory: 234.36 GB
FP16 · 8K context
Base Weights: 229.11 GB
KV Cache: 2.75 GB
Runtime & Graphs: 2.50 GB
Minimum Required Hardware
8x NVIDIA H100 80GB
234.4 / 576.0 GB usable · TP 8
Determine requirements for your specific workload:
Open in VRAM Calculator
Architecture Diagram
Evaluation Benchmarks
Rank
#171
| Benchmark | Score | Rank |
|---|---|---|
StackUnseen | 0.364 | 29 |
Software Engineering | 0.54 | 29 |
Agentic Index | 0.02 | 106 |
Web Development | 1197 | 116 |
Coding Index | 0.31 | 117 |
Intelligence Index | 0.09 | 254 |
Rankings
Overall Rank
#171
Coding Rank
#118
About Devstral 2 123B Instruct
Devstral 2 123B Instruct is an agentic dense 123B parameter software engineering model from Mistral AI. It is optimized for multi-file codebase exploration, editing, and powering coding agents, with a 262k context window.
Technical Specifications
Attention
Attention Structure
Multi-Head Attention
Attention Heads
96
Key-Value Heads
8
Attention Head Dimension
128
Position Embedding
Absolute Position Embedding
RoPE Theta
1,000,000
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Hidden Dimension Size
12,288
Number of Layers
88
FFN Intermediate Size (Dense)
28,672
Tokenizer
Vocabulary Size
131,072
About Devstral 2
Devstral 2 is a family of developer-focused models from Mistral AI, optimized for multi-file software engineering tasks and coding agents.
Other Devstral 2 Models
- No related models available