Which Ollama models support tools
Ollama lists tools among a model's capabilities when it can take a tools array. How to check your version and model, a test call, and the library list.

Resource Highlights
- Check the capability on your own install with ollama show
- Why the answer depends on your Ollama version and the model you pulled
- What the library's Tools filter showed on 18 November 2026, and what it leaves out
An Ollama model supports tools when your Ollama server lists tools among that model's capabilities. The server works this out from the model's prompt template or its built-in parser, so the answer depends on two things: the Ollama version you run and the version of the model you pulled. The model library's Tools filter lists models Ollama labels this way, though not all of them at once.
Checked 18 November 2026 against Ollama's documentation, its model library and its source code at release v0.40.16 (published 9 November 2026). None of the commands below were run for this page: no Ollama install and no model were tested. The expected results come from the documentation and the source, linked at each step.
Check your own install
1. Your Ollama version
ollama --version
curl http://localhost:11434/api/version
ollama --version asks the running server for its version and prints ollama version is followed by it. If the command-line client is a different version, it adds a line starting Warning: client version is (source, cmd/cmd.go). The server's version is the one that matters here, because the server decides the capabilities. The /api/version endpoint returns it as JSON with a version field (API reference).
2. The model's capabilities
ollama show qwen3
curl http://localhost:11434/api/show -d '{"model": "qwen3"}'
Replace qwen3 with the model you pulled; qwen3 is the model Ollama's own tool calling examples use. ollama show prints a Capabilities table when the server reports any (source, cmd/cmd.go). The /api/show endpoint returns the same list in a capabilities array, which the API reference describes as a "List of supported features" (API reference).
Look for tools in that list. That is the exact string Ollama uses for tool support (source, types/model/capability.go). Other values in the same list include completion, thinking, vision and embedding.
3. A single tool call
curl http://localhost:11434/api/chat -d '{
"model": "qwen3",
"stream": false,
"messages": [
{"role": "user", "content": "What is the temperature in New York?"}
],
"tools": [{
"type": "function",
"function": {
"name": "get_temperature",
"description": "Get the current temperature for a city",
"parameters": {
"type": "object",
"required": ["city"],
"properties": {
"city": {"type": "string", "description": "The name of the city"}
}
}
}
}]
}'
This follows the single-tool request in Ollama's tool calling documentation. What to expect:
- If the model has the
toolscapability and decides to use the tool, the reply'smessagecarries atool_callslist, each entry with the function'snameand itsarguments(docs). Your code runs the function and sends the result back as a message with"role": "tool". The model may also answer in plain text without calling the tool; the documentation does not promise a call. - If the model lacks the
toolscapability, the server refuses the request with HTTP 400 and an error that names the model and ends indoes not support tools. A request that includes tools requires the capability (source, server/routes.go), the check fails when it is missing (server/images.go), the model's name is put in front of it (server/routes.go), and the error is returned as a bad request (server/routes.go). The documentation does not describe this error.
How Ollama decides
At v0.40.16 the server adds tools to a model's capabilities in three cases (source, server/images.go):
- the model's Go template uses a
toolsvariable; - the model's chat template mentions
toolsortool_call; - the model uses one of Ollama's built-in parsers that handles tool calls (source).
So the capability is read from files that arrive with the model and from parsers that ship with Ollama. Two consequences follow from that, and both are worth checking yourself: a copy of a model pulled long ago carries the template it was pulled with, and the list of built-in parsers belongs to the Ollama version you run.
What the label leaves open
tools means the server will accept a tools array for that model and look for tool calls in its output. It says nothing about how well the model picks a tool or fills its arguments. That takes a test with your own tools and prompts.
Ollama's documentation also covers streaming tool calls, parallel tool calls and a multi-turn agent loop. For what a tool call is and what can go wrong when language triggers an action, see Tool use: when language triggers actions.
What the library's Tools filter showed on 18 November 2026
The table below is an observed list: what the Tools filter showed when it was read on 18 November 2026. It is incomplete, and it changes with the sort order and over time.
- It shows 20 models and no total. Neither sort order offered a next page or a count.
- It depends on the sort. Sorted by Popular and by Newest, the filter showed 20 models each, and 12 of them appeared in both lists.
- It can leave out tagged models.
qwen3, the model in Ollama's own examples, carries thetoolstag on its library page and appeared in neither list.
To check a model that is not listed, open its library page and look for the tools tag, or run ollama show on your own copy (step 2 above).
Sorted by Popular, in page order, as observed on 18 November 2026:
| Model | Runs | Other capability tags |
|---|---|---|
| deepseek-v4.1-flash | cloud | vision, thinking |
| glm-5.3 | cloud | thinking |
| glm-5.3-flash | cloud | vision, thinking |
| inkling-2-small | cloud | vision, thinking |
| ripple-preview | local | thinking |
| mimo-v2.6 | local | vision, thinking |
| lfm3 | local | thinking |
| lfm3-omni | local | vision, audio |
| lfm3-vl | local | vision, thinking |
| lfm3-audio | local | vision, audio |
| minimax-m3.1-flash | cloud | vision, thinking |
| kimi-k3.1-pro | cloud | vision, thinking |
| qwen3.8 | local | vision, thinking |
| nemotron-3.5-lightning | local | thinking |
| bonsai-2 | local | vision, thinking |
| ornith-1.5 | local | vision, thinking |
| ornith | local | thinking |
| flint | local | thinking |
| muse-glimmer | local | vision, thinking |
| tev1 | local | decision, thinking |
Models marked cloud carry the cloud tag in the library and list no download size. The library tag is Ollama's label for the model; the capability your server reports for the copy you pulled is the one your requests are checked against.
Dated history
- 25 July 2024: Ollama announced tool calling, naming Llama 3.1, Mistral Nemo, Firefunction v2 and Command-R+.
- 28 May 2025: streaming tool calls, with a parser that reads each model's template for the start of a tool call instead of waiting for complete JSON.
- 27 October 2026: release v0.40.12 fixed tool-only assistant turns from Kimi CLI.
- 9 November 2026: v0.40.16, the latest release when this page was checked. A pre-release, v0.40.17-rc1, followed on 10 November.
Neither blog post names the Ollama version that introduced the feature.
Not tested here
This page has no "tested on" table. Nothing was installed or run for it. A tested row would record the Ollama version, the model and tag, the operating system, the exact commands and their output, the date, and who ran it.
Related on the Shelf: Tools: Ollama, Ollama launch and Local LLM references.