Model
Model explorer

Codestral Mamba (7B)

OPEN
Mistral AI · Codestral family · released Jul 16, 2024

First Mamba2-architecture code model from Mistral; linear-time inference, tested to 256K-token in-context retrieval, Apache 2.0.

ReasoningCodingVisionFunction callingTool useAgentic
774.0
Elo · unrated
Parameters
7.3B
Active params
7.3B (dense)
Context
256K tokens
Architecture
Mamba2 state-space model (linear-time inference)
License
Apache 2.0
Languages
—
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
HumanEvalCoding75.0%#64
best: Claude Opus 4.5 · 99.4%
MBPPCoding68.5%#42
best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%
Run it locally
VRAM @ Q4
5 GB
VRAM @ FP16
15 GB
Fits on (Q4)
RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Mamba2 quantization/support in llama.cpp was experimental and unstable at launch; state-space caching differs from transformer KV-cache VRAM scaling
Quantizations
GGUF Q4 (experimental Mamba2 support)
Codestral family
Elo progression across releases
Codestral 22BMay 2024458.9Codestral Mamba (7B)Jul 2024774.0Codestral 25.01Jan 20251056.8
Resources
HFmistralai/Mamba-Codestral-7B-v0.1DOCSdocs.mistral.ai/models/model-cards/codestral-mamba-7b-0-1APIAvailable via Mistral AI API
API price weights · each benchmark row carries its own source badge (see methodology)