Mistral’s Medium line has seen no new general-purpose release since Medium 3.5 launched in late April 2026 as a 128B dense multimodal model unifying chat, reasoning, and coding with a 256k context window. The company has instead prioritized specialized releases and its new flagship, Mistral Large 4, a roughly 1T-parameter MoE model unveiled in public preview on October 6 with open weights slated for late October. This shift reflects Mistral’s strategy of consolidating capabilities into higher-tier models while competitors accelerate iteration, leaving traders focused on whether a 3.6+ Medium emerges before year-end or yields to Large-series dominance and narrower tools like Leanstral or OCR updates. Key near-term catalysts include Large 4’s weight release and any post-training roadmap details that could signal the next Medium checkpoint.