modelgrep

Meta: Llama 3.3 70B Instruct vs Mistral: Ministral 3 8B 2512

Meta: Llama 3.3 70B Instruct wins on more metrics (5 of 9), but the right pick depends on what you optimize for — see the breakdown below.

MetricMeta: Llama 3.3 70B InstructMistral: Ministral 3 8B 2512
Intelligence Index14.514.8
Coding Index10.710.0
GPQA Diamond50%47%
Design Arena Elo1112
Speed (tokens/sec)11022
Latency227ms254ms
Input price /M$0.100$0.150
Output price /M$0.320$0.150
Context window131K262K
CapabilitiesToolsJSONToolsJSONVision