[BUG] Claude Code hangs with local Ollama on a trivial prompt, while direct /v1/messages works

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Claude Code hangs when using a local Ollama backend through Anthropic-compatible configuration.

The problem happens even with a trivial prompt like ciao.

Important detail: the same Ollama model responds correctly when called directly through Ollama's Anthropic-compatible /v1/messages endpoint, so this does not look like a basic Ollama connectivity or basic /v1/messages compatibility problem.

This reproduces with local Ollama models on my setup and does not appear to be specific to a single model.

What Should Happen?

Claude Code should return a normal response for a trivial prompt, just like the same model does when queried directly through Ollama's Anthropic-compatible /v1/messages endpoint.

Error Messages/Logs

Environment:
- Claude Code: 2.1.114
- Ollama: 0.21
- OS: Ubuntu 25.10
- Kernel: 6.17.0-20-generic
- CPU: AMD Ryzen 7 6800H
- CPU cores: 8
- Logical CPUs / threads: 16
- RAM: 30 GiB
- GPU: no discrete GPU
- Graphics hardware present on the machine: AMD Radeon 680M integrated graphics
- Actual inference mode on this setup: CPU-only

Ollama systemd override:
[Service]
User=lupo
Group=lupo
Environment="HOME=/home/lupo"
Environment="OLLAMA_MODELS=/home/lupo/.ollama/models"
Environment="OLLAMA_CONTEXT_LENGTH=65536"
Environment="OLLAMA_NUM_PARALLEL=1"
Environment="OLLAMA_KEEP_ALIVE=1m"
CPUAccounting=yes
CPUQuota=400%
CPUQuotaPeriodSec=10ms
Nice=10
IOSchedulingClass=idle

Claude Code / Anthropic-compatible configuration:
- ANTHROPIC_AUTH_TOKEN=ollama
- ANTHROPIC_BASE_URL=http://127.0.0.1:11434

Main reproduction model:
- qwen2.5-coder:7b-opencode-32k

Direct Ollama Anthropic-compatible request that succeeds:
curl http://127.0.0.1:11434/v1/messages \
  -H 'Content-Type: application/json' \
  -H 'x-api-key: ollama' \
  -H 'anthropic-version: 2023-06-01' \
  -d '{
    "model": "qwen2.5-coder:7b-opencode-32k",
    "max_tokens": 64,
    "messages": [
      {
        "role": "user",
        "content": "ciao"
      }
    ],
    "stream": false
  }'

Response:
{"id":"msg_e1a1726675e5761fd7896619","type":"message","role":"assistant","model":"qwen2.5-coder:7b-opencode-32k","content":[{"type":"text","text":"Ciao! Come posso aiutarti oggi?"}],"stop_reason":"end_turn","usage":{"input_tokens":31,"output_tokens":12}}

Relevant Ollama log line from successful direct test:
[GIN] 2026/04/20 - 19:45:21 | 200 | 7.100566535s | 127.0.0.1 | POST "/v1/messages"

Relevant Ollama log line observed during Claude Code usage:
[GIN] 2026/04/20 - 19:35:59 | 500 | 5m57s | 127.0.0.1 | POST "/v1/messages?beta=true"

Additional successful direct-test model/runtime details:
qwen2.context_length = 32768
llama_context: n_ctx = 32768
time=2026-04-20T19:45:17.491+02:00 level=INFO source=server.go:1402 msg="llama runner started in 3.29 seconds"
[GIN] 2026/04/20 - 19:45:21 | 200 | 7.100566535s | 127.0.0.1 | POST "/v1/messages"

Steps to Reproduce

1. Start Ollama locally on 127.0.0.1:11434

1. Configure Claude Code to use Ollama through:

• ANTHROPIC_AUTH_TOKEN=ollama
• ANTHROPIC_BASE_URL=http://127.0.0.1:11434

1. Launch Claude Code with a local model, for example:

claude --model qwen2.5-coder:7b-opencode-32k

1. Send a trivial prompt:

ciao

1. Observe that Claude Code hangs and does not return a response in the UI.

1. Compare this with a direct Ollama Anthropic-compatible request to /v1/messages using the same model, which succeeds and returns a normal response.

Claude Model

Other

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.114

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

_No response_