modelgrep

Meta: Llama 3.1 8B Instruct vs Mistral: Ministral 3 3B 2512

MMeta: Llama 3.1 8B Instructmeta-llama/llama-3.1-8b-instructVSMMistral: Ministral 3 3B 2512mistralai/ministral-3b-2512

Meta: Llama 3.1 8B Instruct wins on more metrics (6 of 8), but the right pick depends on what you optimize for — see the breakdown below.

MetricMeta: Llama 3.1 8B InstructMistral: Ministral 3 3B 2512
Intelligence Index11.8✓11.2
Coding Index4.9✓4.8
GPQA Diamond26%36%✓
Design Arena Elo—1072✓
Speed (tokens/sec)171✓45
Latency151ms✓253ms
Input price /M$0.020✓$0.100
Output price /M$0.030✓$0.100
Context window131K131K
CapabilitiesToolsJSONToolsJSONVision
Meta: Llama 3.1 8B Instruct details →Mistral: Ministral 3 3B 2512 details →