Atlas

Models

← All models

Ministral 8B Instruct 2410

Mistral AI · 2024-10-16 · 8B parameters

Ministral 8B Instruct 2410 is the instruction-tuned October 2024 release of Mistral AI's Ministral 8B, part of the Ministraux edge-deployment family. It is fine-tuned for chat, instruction following, and tool use within an 8B footprint supporting a 128K-token context window and interleaved full and sliding-window attention for efficient on-device inference. Available on Hugging Face for research use, it is suitable for local and privacy-sensitive deployments.

Benchmark scores

BenchmarkScore
BBH (Open LLM Leaderboard v2)47.6
BFCL v4 Overall Accuracy11.1
BRIDGE Medical (chain-of-thought)25.9
BRIDGE Medical (few-shot)39.6
BRIDGE Medical (zero-shot)30.4
BuzzBench32.3
Capability109.3
GPQA (Open LLM Leaderboard v2)28.4
IFEval (Open LLM Leaderboard v2)59.0
MATH Level 5 (Open LLM Leaderboard v2)19.6
MMLU-Pro (Open LLM Leaderboard v2)32.9
MuSR (Open LLM Leaderboard v2)41.4
SnakeBench Average Apples0.5
SnakeBench Best Apples4.0
SnakeBench Rating18.7
SnakeBench Total Apples15.0
SnakeBench Win Rate41.9
Loading Atlas data…