Ministral 3 14B 2512

Ministral 3 14B 2512 is a compact open-weight release from Mistral AI that fits into the broader Ministral family of efficient models aimed at workloads where a smaller footprint matters more than raw frontier scale. It is positioned on third-party leaderboards alongside much larger contemporaries such as Claude Fable 5.1, Claude Opus 5, Kimi K3, Qwen3.8 2.4T A95B, and GLM 5.3 variants, which signals that Mistral is benchmarking this 14B-class entry against both flagship and compact peers rather than treating it as an internal baseline. Creator attribution is consistently labeled to Mistral AI across aggregator and routing listings, keeping the model clearly separated from any serving or gateway providers.

The model's intended use centers on cost-sensitive, high-throughput applications where predictable per-token pricing and open-weight flexibility are priorities. A behavioral evaluation row on the Vectara HHEM Hallucination Leaderboard places Ministral 3 14B 2512 near a median sixth-percentile position, giving a qualitative read on its summarization faithfulness relative to larger frontier peers without overstating a numeric rank. Aggregator pricing of 0.20 dollars per million input and output tokens makes it economically attractive for production pipelines that need to balance capability against spend, and the open-weight posture allows teams to self-host or fine-tune for specialized assistants, document understanding, and structured-output tooling rather than relying solely on a hosted endpoint.