Atlas

Models

← All models

Mistral 7B v0.1 Base

Mistral AI · 2023-09-27 · 7.2B parameters

Mistral 7B v0.1 Base is Mistral AI's first open pretrained model, released on September 27, 2023. The Apache-2.0 checkpoint has an 8,192-token context length, grouped-query attention, and a 4,096-token sliding-attention window whose stacked layers propagate information beyond each local window; Mistral lists the model as retired on March 30, 2025.

Benchmark scores

BenchmarkScore
Adversarial NLI47.1
BBH (Open LLM Leaderboard v2)44.2
Capability104.0
Epoch Capabilities Index (ECI)111
Global PIQA - Non-Parallel (Log-Likelihood Accuracy)53.3
Global PIQA - Parallel (Log-Likelihood Accuracy)23.5
GPQA (Open LLM Leaderboard v2)29.2
HellaSwag (Unspecified Scoring Protocol)81.0
IFEval (Open LLM Leaderboard v2)23.9
LegalBench53.8
LegalBench - Conclusion Tasks61.6
LegalBench - Interpretation Tasks49.6
LegalBench - Issue Tasks51.7
LegalBench - Rhetoric Tasks51.1
LegalBench - Rule Tasks54.8
MATH Level 5 (Open LLM Leaderboard v2)2.9
MedMCQA (Open Medical-LLM)48.2
MedQA 4-options (Open Medical-LLM)50.8
MMLU Anatomy (Open Medical-LLM)55.6
MMLU Clinical Knowledge (Open Medical-LLM)68.7
MMLU College Biology (Open Medical-LLM)68.1
MMLU College Medicine (Open Medical-LLM)59.5
MMLU Medical Genetics (Open Medical-LLM)71.0
MMLU Professional Medicine (Open Medical-LLM)68.4
MMLU-Pro (Open LLM Leaderboard v2)30.1
MuSR (Open LLM Leaderboard v2)41.4
OpenBookQA79.8
PubMedQA (Open Medical-LLM)75.4
Loading Atlas data…