This repository provides ik_llama optimized GGUF quantizations for the mistralai/Mistral-Small-4-119B-2603 model.
Optimization and Importance Matrix
These files were generated using the ik_llama.cpp fork, which offers row interleaved formats (R4) specifically designed to minimize degradation in MoE architectures. I send my biggest thanks to the author ikawrakow for developing and maintaining this tool.
To maximize precision, an importance matrix (iMatrix) was computed prior to quantization. This calculation was performed on a RunPod infrastructure equipped with 3 NVIDIA A100 80GB accelerators, ingesting the calibration_data_v5_rc.txt dataset provided by tristandruyen.
Perplexity Scores
All tests were performed on the wikitext-2 dataset using a context window of 1024 tokens and a batch size of 512.
Model Format
Perplexity (PPL)
Size (GiB)
Bits per Weight (BPW)
BF16 (Unsloth Base)
5.3035
221.64
16.00
IQ5_K_R4 (ik_llama)
5.3431
82
5.51
UD_Q4_K_S (Unsloth)
5.3883
69.30
~ 4.50
IQ4_K_R4 (ik_llama)
5.4242
67.2
4.31
Vision and Multimodality Note
This model features vision capabilities. You may find the original model's .mmproj file in the "Files and versions" tab.
================== Original Description ==================
Mistral Small 4 is a powerful hybrid model capable of acting as both a general instruction model and a reasoning model. It unifies the capabilities of three different model families—Instruct, Reasoning (previously called Magistral), and Devstral—into a single, unified model.
With its multimodal capabilities, efficient architecture, and flexible mode switching, it is a powerful general-purpose model for any task. In a latency-optimized setup, Mistral Small 4 achieves a 40% reduction in end-to-end completion time, and in a throughput-optimized setup, it handles 3x more requests per second compared to Mistral Small 3.
To further improve efficiency you can either take advantages of:
Mistral Small 4 includes the following architectural choices:
MoE: 128 experts, 4 active.
119B parameters, with 6.5B activated per token.
256k context length.
Multimodal input: Accepts both text and image input, with text output.
Instruct and Reasoning functionalities with function calls (reasoning effort configurable per request).
Mistral Small 4 offers the following capabilities:
Reasoning Mode: Toggle between fast instant reply mode and reasoning mode, boosting performance with test-time compute when requested.
Vision: Analyzes images and provides insights based on visual content, in addition to text.
Multilingual: Supports dozens of languages, including English, French, Spanish, German, Italian, Portuguese, Dutch, Chinese, Japanese, Korean, and Arabic.
System Prompt: Strong adherence and support for system prompts.
Agentic: Best-in-class agentic capabilities with native function calling and JSON output.
Speed-Optimized: Delivers best-in-class performance and speed.
Apache 2.0 License: Open-source license for both commercial and non-commercial use.
Large Context Window: Supports a 256k context window.
Recommended Settings
Reasoning Effort:
'none' → Do not use reasoning
'high' → Use reasoning (recommended for complex prompts)
Use reasoning_effort="high" for complex tasks
Temperature: 0.7 for reasoning_effort="high". Temp between 0.0 and 0.7 for reasoning_effort="none" depending on task.
Use Cases
Mistral Small 4 is designed for general chat assistants, coding, agentic tasks, and reasoning tasks (with reasoning mode toggled). Its multimodal capabilities also enable document and image understanding for data extraction and analysis.
Its capabilities are ideal for:
Developers interested in coding and agentic capabilities for SWE automation and codebase exploration.
Enterprises seeking general chat assistants, agents, and document understanding.
Researchers leveraging its math and research capabilities.
Mistral Small 4 is also well-suited for customization and fine-tuning for more specialized tasks.
Examples
General chat assistant
Document parsing and extraction
Coding agent
Research assistant
Customization & fine-tuning
And more...
Benchmarks
Comparison with internal models
Depending on your tasks you can trigger reasoning thanks to the support of the per-request parameter reasoning_effort. Set it to:
reasoning_effort="high": Deep, step-by-step reasoning for complex problems, with equivalent verbosity to previous Magistral models such as mistralai/Magistral-Small-2509.
Internal benchmark
Comparing Reasoning Models
Internal benchmark - Reasoning
Comparison with other models
Mistral Small 4 with reasoning achieves competitive scores, matching or surpassing GPT-OSS 120B across all three benchmarks while generating significantly
shorter outputs. On AA LCR, Mistral Small 4 scores 0.72 with just 1.6K characters, whereas Qwen models require 3.5-4x more output (5.8-6.1K)
for comparable performance. On LiveCodeBench, Mistral Small 4 outperforms GPT-OSS 120B while producing 20% less output.
This efficiency reduces latency, inference costs, and improves user experience.
Comparison benchmark - LCR
Comparison benchmark - LiveCodeBench
Comparison benchmark - AIME25
Usage
You can find Mistral Small 4 support on multiple libraries for inference and fine-tuning. We here thank everyone contributors and maintainers that helped us making it happen.
