Skip to content
CC Switch WikiUnofficial Community Wiki

Claude Code Tool Call Failure via NVIDIA Proxy - CC Switch Fix

Quick fix

NVIDIA free-tier models are unstable; switch to stable providers like deepseek-ai/deepseek-v4-flash.

Symptom

Section titled “Symptom”
Error output
他无法正常调用工具
English translation

He cannot call tools normally

Cause

Section titled “Cause”

When using the NVIDIA free tier proxy via cc-switch, multiple models (e.g., glm series, certain qwen versions) suffer from severe connection timeouts or streaming response anomalies. Real-world data shows some models have first-byte wait times of tens of seconds or direct TimeoutError, preventing Claude Code from receiving valid tool_calls responses. Additionally, some models (e.g., kimi series) return results but with formats non-compliant with OpenAI standards, causing parsing failures. This is not a bug in cc-switch itself, but rather upstream rate limits (TPM/TPD) or network instability from free providers. Although PR #2557 attempted to fix related streaming parsing issues, the root cause remains provider-side instability.

Notably, even if using better-performing deepseek models, tool call failures may still occur due to provider quotas or temporary network fluctuations. The community-confirmed stable alternative is switching to other reliable providers.

Fix

Section titled “Fix”

  1. Switch the provider in cc-switch configuration from nvidia to deepseek or another stable provider (e.g., openai/gpt-oss-120b).

  2. If you must use the NVIDIA channel, ensure the selected model is known to be stable (e.g., deepseek-ai/deepseek-v4-flash or deepseek-ai/deepseek-v4-pro), and avoid models prone to timeouts such as glm or qwen.

Affected Versions

Section titled “Affected Versions”
ToolClaude Code
Version3.14.1
PlatformsmacOSLinux

Source Issues

Section titled “Source Issues”

This page is distilled from 2 real issues

FAQ

Section titled “FAQ”
Why can't I call tools even after using deepseek-v4-flash?
Although deepseek-v4-flash performs stably in benchmarks, if your account triggers the provider's rate limits (TPM/TPD) or if there are network fluctuations in your current environment, calls may still fail. Check the provider status or try restarting the cc-switch proxy.
What should I do if clicking the enable proxy button has no effect on Linux?
This is a known UI interaction issue with cc-switch on Linux platforms, which typically does not affect the underlying proxy functionality. You can manually check the proxy settings in the configuration file or start the proxy service via the command line.
Does PR #2557 resolve this issue?
PR #2557 addressed some streaming parsing and connection issues, but the instability of the NVIDIA free channel primarily stems from upstream providers. Even after applying this PR, some models may still experience timeouts or format errors, so the official recommendation is to prioritize more stable providers.
Section titled “Related problems”
cc-switch DeepSeek usage statistics inaccurate / balance query failsUpgrade to v3.19.2 and rebuild Codex usage. The balance query needs a correct {{apiKey}}, or a proxy.17 issues mergedANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEYOpen `~/.claude/settings.json`, delete the duplicate auth variable so only the one your provider needs remains, save, and restart Claude Code.4 issues mergedapi failed deserialize json body intoDepending on the error: roll Claude Code back, upgrade cc-switch, turn the local proxy off, or stop sending unsupported fields.7 issues merged

This is an unofficial community wiki with no affiliation to the cc-switch authors or the project itself. Its content is compiled from the project's public GitHub issues. This site distributes no software.