First model release of 2025 for French AI lab Mistral, who describe Mistral Small 3 as "a latency-optimized 24B-parameter model released under the Apache 2.0 license." More notably, they claim …