Ollama v0.40.15

Summary

Ollama 0.40.15 (2026-11-06) adds Xiaomi's MiMo V2.6 coding model and a family of "AutoTrust" decision models. It also adds Kimi Code CLI to ollama launch and lets recurrent and vision models serve concurrent requests. Running ollama with no arguments now opens a launcher.

Key points

• MiMo V2.6: Xiaomi's 9B model for code and agent workflows: ollama run mimo-v2.6. Licence and benchmarks were not in the release notes.
• AutoTrust decision models: answer questions about text with choices, yes/no probabilities, or ratings. Sizes are 9b, 35b, and 27b-vl; the 35b and 27b-vl also read images. The notes say the autotrust-jev variant runs only on Apple silicon.
• ollama launch for Kimi Code CLI: supports the successor to kimi-cli and offers to migrate existing kimi-cli setups. ollama launch also fills in Pi's model list, and Claude Desktop's model picker shows Ollama model names.
• Serving: recurrent and vision GGUF models handle more than one request at a time. Qwen3 and Qwen3.5 models support YaRN long-context scaling.

Why it matters

A small local coding model and local classifier-style models are now one command away. Concurrent requests make recurrent and vision models usable behind several agents at once.