Ollama 0.40.4: `'nimble:latest' was downloaded in a format that is no longer supported` for decision models (nimble, tev, clef-flash) on `/v1/systemone`

Context

Source(s): https://github.com/ollama/ollama/issues/18998 Status: reported working by source, not reproduced by us.

Problem

After upgrading Ollama from 0.40.3 to 0.40.4, requests to /v1/systemone with the decision models nimble, tev and clef-flash fail with the error above. Only nimble-V3 kept working (our inference: it was probably already published in the new format). Running the suggested ollama pull once did not fix it for the original poster.

Error (redacted / normalized)

{"error":"'nimble:latest' was downloaded in a format that is no longer supported. To re-download, run 'ollama pull nimble:latest'"}

Applies when

Ollama 0.40.4 on Linux (Ubuntu 22.04, NVIDIA Jetson AGX Orin), models pulled before the upgrade.

Confirm it is this problem

1. ollama -v → expect: 0.40.4 or later.
2. ollama list <model>:latest → expect: the ID differs from the ID shown for that tag on the model's ollama.com library page.

Not this when / boundary

The same message for a model whose ollama list ID already matches the library (then something else is wrong; this card does not cover it).

Root cause

An Ollama collaborator explained that before 0.40.4, systemone used a custom engine with models configured for it; 0.40.4 switched to native llama.cpp support, so the models were republished for the new engine and 0.40.4 shows this message when an old copy is used.

Fix

1. Run ollama pull <model>:latest for each affected model (nimble, tev, clef-flash). → expect: ollama list shows a new ID that matches the library.
2. If the ID did not change, pull again later and check the ID again. The original poster's first pull (about 13 hours earlier) kept the old ID; pulling again after the collaborator pointed out the mismatch gave a new ID and the model worked. Why the first pull didn't update is not explained in the thread.

How to verify

Send the same /v1/systemone request → expect: an answers object (for example "type": "noul" with a probability) instead of the error; ollama ps shows the model on the llamacpp runner. The original poster confirmed nimble, clef-flash and tev all worked on 0.40.4 after re-pulling.

What did not work

The first ollama pull that kept the old model ID.

Risks

none known. The collaborator said nothing else is needed after the models are updated.

Verified scope

Ollama 0.40.4 and later with systemone decision models pulled before 0.40.4, October 2026.

Becomes obsolete when

n/a (only affects copies pulled before 0.40.4). Open question: whether you need to re-pull these models on every Ollama upgrade (the original poster asked; the collaborator answered only for this change).