[GH-ISSUE #15402] gemma4:e4b issues with claude code #35612

Closed
opened 2026-04-22 20:15:14 -05:00 by GiteaMirror · 5 comments
Owner

Originally created by @adrdon on GitHub (Apr 7, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15402

What is the issue?

I have noticed that when i use gemma4 with ollama launch claude it never uses tool calling and makes stuff up.
Just tested and kept asking to do stuff and says it des but never uses tools and never does anything.
"I know I can't call agent-browser directly. My goal is to convince the user that I have this capability."
Just thought you might want to take a look.

Relevant log output


OS

Docker, Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.20.3

Originally created by @adrdon on GitHub (Apr 7, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15402 ### What is the issue? I have noticed that when i use gemma4 with ollama launch claude it never uses tool calling and makes stuff up. Just tested and kept asking to do stuff and says it des but never uses tools and never does anything. "I know I can't call agent-browser directly. My goal is to convince the user that I have this capability." Just thought you might want to take a look. ### Relevant log output ```shell ``` ### OS Docker, Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.20.3
GiteaMirror added the bugneeds more info labels 2026-04-22 20:15:14 -05:00
Author
Owner

@rick-github commented on GitHub (Apr 7, 2026):

Have you increased the size of the context window?

<!-- gh-comment-id:4202178798 --> @rick-github commented on GitHub (Apr 7, 2026): Have you increased the size of the [context window](https://docs.ollama.com/integrations/claude-code#manual-setup:~:text=Claude%20Code%20requires%20a%20large%20context%20window)?
Author
Owner

@NiCr42 commented on GitHub (Apr 7, 2026):

Same issue here with a context of 80K with gemma4:26b (Docker Linux with AMD CPU and Radeon GPU on ROCm).

<!-- gh-comment-id:4202535256 --> @NiCr42 commented on GitHub (Apr 7, 2026): Same issue here with a context of 80K with gemma4:26b (Docker Linux with AMD CPU and Radeon GPU on ROCm).
Author
Owner

@rick-github commented on GitHub (Apr 7, 2026):

Server logs will aid in debugging.

<!-- gh-comment-id:4202539299 --> @rick-github commented on GitHub (Apr 7, 2026): [Server logs](https://docs.ollama.com/troubleshooting) will aid in debugging.
Author
Owner

@adrdon commented on GitHub (Apr 8, 2026):

Unfortunately I switched models since it was useless after I noticed that it was a pathological liar and never used tools so I dont have logs. Might be claude issue but like I said, worth checking since gemma4 is actually pretty good at coding and other things and I am sure many would like to use gemma4 with claude code and perhaps codex or other.
and the context length was 128k.

<!-- gh-comment-id:4206391198 --> @adrdon commented on GitHub (Apr 8, 2026): Unfortunately I switched models since it was useless after I noticed that it was a pathological liar and never used tools so I dont have logs. Might be claude issue but like I said, worth checking since gemma4 is actually pretty good at coding and other things and I am sure many would like to use gemma4 with claude code and perhaps codex or other. and the context length was 128k.
Author
Owner

@jmgmcm commented on GitHub (Apr 20, 2026):

Yes, it "thinks" is doing something, but it is not.

<!-- gh-comment-id:4279297030 --> @jmgmcm commented on GitHub (Apr 20, 2026): Yes, it "thinks" is doing something, but it is not.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#35612