Error: pull model manifest: file does not exist means the registry has no model under the name and tag you typed. Nothing was downloaded and nothing on your machine is broken; copy the exact tag from the model's page in the Ollama library and pull again.

The tag after the colon is the part that usually goes wrong: a size or variant that was never published, a typo, or a name from a blog post written for another model family. Pull a tag that exists, such as qwen2.5:3b, and check the result with ollama list.

The error

PS> ollama pull llama3:70b-nonexistent
pulling manifest
Error: pull model manifest: file does not exist

Why it happens

A model in Ollama is addressed as name:tag. Before it fetches any layer, ollama pull asks the registry for the manifest of that exact pair, the small file that lists the layers to download. When the registry has no manifest for the pair, the client stops at the "pulling manifest" step and prints this message. The name alone being right is not enough: llama3 exists, the tag 70b-nonexistent does not, so the whole reference fails.

The progress spinner in the terminal only ever shows "pulling manifest", so the message reads like a disk or permission problem. It is not; the command exits with code 1 before it writes anything to your models folder.

The fix

PS> ollama pull qwen2.5:3b
pulling manifest
pulling 5ee4f07cdb9b: 100% ▕██████████████████▏ 1.9 GB
pulling 66b9ea09bd5b: 100% ▕██████████████████▏   68 B
pulling eb4402837c78: 100% ▕██████████████████▏ 1.5 KB
pulling b5c0e5cf74cf: 100% ▕██████████████████▏ 7.4 KB
pulling 161ddde4c9cd: 100% ▕██████████████████▏  487 B
verifying sha256 digest
writing manifest
success

PS> ollama list
NAME                       ID              SIZE      MODIFIED
qwen2.5:3b                 357c53fb659c    1.9 GB    Less than a second ago
llama3.1:8b                46e0c10c039e    4.9 GB    3 weeks ago

Open the model's page on ollama.com, switch to its tags view and copy the full name:tag string from there instead of typing it. In this run the layers were already on disk, so the pull finished in about a second and every layer showed 100% at once; on a fresh machine the same command downloads 1.9 GB. If you already have a model and only want to know its exact tag, ollama list prints it in the NAME column, and that string is what your code and your ollama run commands must use.

How it was reproduced

On Windows 11 with Ollama 0.40.1 running as the tray app, in Windows PowerShell 5.1, ollama pull llama3:70b-nonexistent printed the error above and exited with code 1. The fix was confirmed with ollama pull qwen2.5:3b (exit code 0) and ollama list; the list is shortened to two of its five rows. The tag 70b-nonexistent was invented for the test so that the request would miss.

Frequently asked

What does 'pull model manifest: file does not exist' mean in Ollama?
The registry has no model for the exact name and tag you asked for. Usually the tag is misspelled or was never published. Nothing was downloaded and your installation is fine.
How do I find the correct Ollama model tag?
Open the model's page in the Ollama library, look at its tags and copy the full name:tag string. For models already on your machine, ollama list shows the exact names in the NAME column.
Does a failed ollama pull leave partial files behind?
Not for this error. The client stops while asking for the manifest, before any layer is downloaded, and exits with code 1.

More decoded errors in the Fixes category. If the pull fails before it even reaches the manifest step on Linux, see could not connect to ollama server.