Ollama v0.40.15 Released with Parallel Vision GGUF Support

Ollama v0.40.15 Released with Parallel Vision GGUF Support

At a Glance

Item Value
Repository ollama/ollama
Version v0.40.15
Published 2026-11-04
License MIT
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

Overview

Ollama (MIT license), a Go-based tool for running and managing various open-weight models locally, has released its latest version, “v0.40.15".

The most impactful change for users in this version is the support for concurrent parallel processing of multiple requests in recurrent models and vision GGUF models. This improves throughput when calling image recognition and other models in parallel from multiple clients.

Key Changes

Concurrent Multi-Request Processing for Recurrent and Vision GGUF Models

Recurrent models and vision GGUF models can now process multiple requests simultaneously.

This affects users operating vision models or recurrent models in parallel via Web APIs or similar setups. No configuration changes or re-downloading of models are required; updating Ollama to this version enables the feature immediately.

Enhanced Third-Party Tool Integration (Kimi Code CLI, Pi, Claude Desktop)

Support for external tools and desktop apps linkable via the ollama launch command has been improved.

It now supports Kimi Code CLI, the successor to kimi-cli, displaying guidance to migrate existing kimi-cli settings. Additionally, Pi’s model list is now automatically populated so models can be selected within Pi, and the Claude Desktop model selection screen has been revised to display actual Ollama model names. This affects users combining these tools with Ollama.

Long-Context (YaRN) Scaling Support for Qwen3 and Qwen3.5

Inference is now supported for Qwen3 and Qwen3.5 models that use YaRN (Yet Another RoPE for Transformers) context scaling.

This affects users leveraging the Qwen3 series for long-context prompt processing or document analysis. It becomes available by running compatible models after updating to the latest version.

Support for New Models mimo-v2.6 and autotrust-jev

mimo-v2.6, a 9B model developed by Xiaomi for coding and agent workflows, is now supported and available via ollama run mimo-v2.6.

Additionally, autotrust-jev, a group of decision-making models that return choices, yes/no probabilities, and evaluations in response to provided questions, has been added. Available sizes include 9b, 35b, and 27b-vl, with the 35b and 27b-vl variants also supporting image inputs. Note that in this release, autotrust-jev is supported on Apple silicon environments.

Welcome Launcher Display on Argument-Free Execution

Running the ollama command by itself without specifying options or subcommands now opens a welcome launcher.

This affects all users who launch Ollama directly from the CLI.

Supported Models and Hardware

This version introduces support for new model architectures and optimizations for specific hardware environments.

Support for Xiaomi “mimo-v2.6"

The “mimo-v2.6" model (9B parameters), developed by Xiaomi for code generation and agent workflows, is newly supported. It can be used when you want to execute programming assistance or agent tasks. You can fetch and run the model directly using the following command:

ollama run mimo-v2.6

Support for Decision-Making Model “autotrust-jev" Series

Support has been added for the decision-making model “autotrust-jev", which reads input text or images and outputs choices, yes/no determination probabilities, and evaluation ratings for given questions.

This model series offers the following multiple model sizes and types:

  • 9b model: A basic-sized decision-making model specialized for text input.
  • 35b model: A larger model capable of more advanced judgments, supporting image reading in addition to text.
  • 27b-vl model: A vision-language (VL) model equipped with image input analysis capabilities.

When running, invoke it by specifying a question with the --ask option along with the text to evaluate. For example, to evaluate a double-charging inquiry, run it with a command like the following:

ollama run autotrust-jev "A customer says their card was charged twice." --ask "Is this a billing issue?"

Note that in this release, autotrust-jev is supported for operation on Apple silicon (Macs with M-series processors).

Qwen3 and Qwen3.5 Support with Long-Context (YaRN) Scaling

Models with Qwen3 and Qwen3.5 architectures using YaRN (Yet Another RoPE for Transformers) scaling technology are now supported. This allows GGUF models of the Qwen3 series, which support long contexts, to process and extend context lengths normally and efficiently during inference. This is a useful feature enhancement for users analyzing massive codebases or reading long texts.

How to Get It

Ollama installation and update procedures vary depending on your platform and environment. For instructions on updating to this release (v0.40.15), please follow the guidance provided in the official documentation and the GitHub releases page.

For procedures on fresh installations or updating existing environments, please refer to the following official release page and library guide:

Releases Since Our Last Article

Compiled by Local Model Watch from the project’s GitHub releases: the versions between this release and the last one we covered, which did not get separate articles. Full history: release tracker.

Version Released Release notes
v0.40.14 2026-11-03 GitHub
v0.40.13 2026-10-29 GitHub

Related Articles

What to Read Next

Sources