Qwen3‑235B‑A22B‑Instruct‑2507 now runs fully local through @EternalAI_ and is stored on Filecoin. It skips thought chains, outputs direct responses, and performs well on reasoning, math, code, and multilingual tasks. The model supports a 256K context window.