Mistral Small 4 · 119B

Mistral Small 4 119B 2603 is an Apache-2.0 licensed model from the mistralai repository, tagged with safetensors. Its official model card describes it as a hybrid model for general instruction and reasoning that unifies Instruct, Reasoning (formerly Magistral), and Devstral capabilities. It is a sparse MoE model with 119B parameters and 6.5B activated per token, 128 experts with 4 active, a 256k context length, multimodal text-and-image input with text output, multilingual support, system-prompt support, and agentic capabilities with native function calling and JSON output. The card also describes Instruct and Reasoning modes, configurable reasoning effort, reasoning and vision capabilities, and Apache-2.0 licensing.

Verified sourceRepository opened Jan 23, 2026Source checked 9/23/2026Version: a11f36be

LOCALRENTED GPUOPEN WEIGHTS

At a glance

Parameters: 119B total · 6.5B active
Architecture: Mixture of experts
Context length: 262,144
License: Apache 2.0
Software: vLLM

View the model source

Sources and files

Publisher model card and vLLM instructions

MY HARDWARE

Will it run on your machine?

Save your machine to see a personalized rating and its reasoning.

Add my hardware

Best for

General chat assistant
Document parsing and extraction
Coding and agentic tasks
Reasoning tasks with reasoning mode toggled
Document and image understanding for data extraction and analysis

Tradeoffs

The source does not provide details on hardware requirements or deployment constraints.
Quantization, speculative decoding, and performance figures are described only as supplied; no independent verification is provided.
The model card focuses on text output; it does not describe other output modalities.
The source lists supported languages but does not provide compatibility or performance details for each language or platform.
No safety, bias, or failure-mode evaluation details are supplied.

Ways to run it

vLLM · Linux

Repository-specific example documented in the publisher model card; confirm dependencies and platform in the source.

vllm serve mistralai/Mistral-Small-4-119B-2603 --max-model-len 262144 --tensor-parallel-size 2 --attention-backend FLASH_ATTN_MLA  --tool-call-parser mistral --enable-auto-tool-choice --reasoning-parser mistral --max_num_batched_tokens 16384 --max_num_seqs 128  --gpu_memory_utilization 0.8

Official instructions