Mistral Large 2 (Instruct 2407)

Name: Mistral Large 2 (Instruct 2407)
Author: Mistral AI

Mistral AI · 2024-07-24 · 122.6B parameters

Mistral Large 2 is Mistral AI's July 2024 Mistral-Large-Instruct-2407 checkpoint, a 123B dense instruction-tuned language model with a 128K-token context window. It improved coding, mathematics, reasoning, and multilingual support, and its weights were released under the Mistral Research License.

Benchmark scores

Benchmark	Score
AA-LCR	1.7
AA-LCR v1.1	2.0
Agentic Tool Use (Chat)	40.4
Agentic Tool Use (Enterprise)	50.4
AIME 2025 (Artificial Analysis avg@10)	0.0
Arabic	970
Artificial Analysis Progress Index v1	155
BigCodeBench-Hard Complete	29.7
BigCodeBench-Hard Instruct	22.3
BuseyBench SVG	2.5
Capability	122.0
Chinese	1006
Coding	1029
Confabulations Leaderboard Confabulation Rate	32.2
Confabulations Leaderboard Non-Response Rate	10.6
Confabulations Leaderboard Weighted Score	21.4
Creative Writing v2	77.2
DTBench	61.2
Elimination Game TrueSkill Mu	4.1
Epoch Capabilities Index (ECI)	128
EQ-Bench v2	85.0
EQ-Bench v2 + MAGI-Hard Combined	78.7
Eval Connections	5.0
Eval Connections — Guess Accuracy	9.1
ForecastBench Baseline Dataset Brier Index	56.0
ForecastBench Baseline Market Brier Index	56.0
ForecastBench Baseline Overall Brier Index	56.0
ForecastBench Preliminary Dataset Brier Index	55.9
ForecastBench Tournament Dataset Brier Index	55.9
ForecastBench Tournament Market Brier Index	68.6