Ollama v0.40.15 Adds Xiaomi MiMo V2.6, AutoTrust Decision Models, and Multi-Request GGUF Serving
What changed
Ollama has released version 0.40.15, introducing support for new model architectures and expanding local execution capabilities:
- Xiaomi MiMo V2.6: Added support for
mimo-v2.6, a 9-billion parameter model tailored for coding tasks and AI agent workflows. - AutoTrust Decision Models: Integrated the
autotrust-jevmodel family designed for direct decision evaluation. These models analyze text or images to return probabilities, ratings, and explicit choices using flags like--ask. The model sizes include 9b, 35b, and 27b-vl (with the 35b and 27b-vl variants supporting visual inputs). In this release,autotrust-jevis supported on Apple silicon. - Engine & Concurrency Updates: Vision and recurrent GGUF models can now handle multiple concurrent requests simultaneously. The update also adds support for long-context scaling (YaRN) in Qwen3 and Qwen3.5 models.
- Developer Workflows & Integrations: The
ollama launchcommand now supports Kimi Code CLI (the successor tokimi-cli) with automatic migration prompts for existing environments. It also automatically populates Pi’s model selector and updates Claude Desktop to display exact Ollama model names. - CLI Behavior: Running the
ollamacommand without arguments now defaults to launching the interactive welcome screen.
Why it matters
Concurrent request execution for vision and recurrent GGUF models allows local runtime environments to handle multi-agent queries or parallel pipeline steps without queuing bottlenecks. Adding dedicated decision models like AutoTrust provides structured binary and rated outputs without relying on complex JSON schemas or heavy prompt engineering.
Furthermore, precise model naming inside third-party interfaces like Claude Desktop and Pi simplifies hybrid local-and-cloud development setups. Practitioners balancing local runtimes with external API surfaces, such as those detailed in our guide to architecting with Claude, gain clearer visibility into local model selection.
Limitations
Support for the autotrust-jev model family in version 0.40.15 is specifically noted for Apple silicon hardware. Additionally, multi-request concurrency in this release is restricted to recurrent and vision-based GGUF formats.
What to do next
To run Xiaomi’s new code and agent model, pull and execute it via the CLI:
ollama run mimo-v2.6On Apple silicon hardware, test decision evaluation with AutoTrust by passing standard queries and an explicit question flag:
ollama run autotrust-jev "A customer says their card was charged twice." --ask "Is this a billing issue?"Existing users running kimi-cli should execute ollama launch to trigger the automatic migration prompt to Kimi Code CLI.
Sources
Share Article
If this article helped you, please share it with others!