Mistral Small 4, released in mid-March 2026 as a 119B-parameter MoE model with only ~6B active parameters per token, unified instruct, reasoning (Magistral), multimodal (Pixtral), and agentic coding (Devstral) capabilities under an Apache 2.0 license, delivering strong efficiency gains like 40% lower latency versus prior Small versions. As of mid-October 2026, no Small 4.1 or higher has been announced, with Mistral prioritizing the Large 4 frontier model (public preview October 6, open weights expected late October) amid competition from efficient open-weight rivals like Qwen and DeepSeek. Traders track Mistral’s roughly quarterly cadence for Small-family updates, API deprecation patterns, and any signals on next hybrid releases that could shift market-implied odds on timing or origin.