date: 2025-07-30 draft: false title: Imperfect combination of Claude Code and Silicon Flow lang: en
** Note: Currently, only the following non-thinking models are supported (more models will be supported later)**
Pro/moonshotai/Kimi-K2-Instructmoonshotai/Kimi-K2-InstructPro/deepseek-ai/DeepSeek-V3deepseek-ai/DeepSeek-V3moonshotai/Kimi-Dev-72Bbaidu/ERNIE-4.5-300B-A47B
date: 2025-07-30 draft: false title: Imperfect combination of Claude Code and Silicon Flow lang: en
1. Pro/moonshotai/Kimi-K2-Instruct
Everything starts well until it outputs to this point:
First initialize the infrastructure of the Node.js project:
null<|tool_call_begin|>functions.Bash:1<|tool_call_argument_begin|>"{command: npm init -y, description: initialize Node.js project package.json}"<|tool_call_end|><|tool_calls_section_end|><|tool_calls_section_end|>
I understand that there was a mistake in calling the tool?
date: 2025-07-30 draft: false title: Imperfect combination of Claude Code and Silicon Flow lang: en
3. deepseek-ai/DeepSeek-V3
A perfect output of what it is going to do. But when I asked it to follow its plan, I repeated what I had just said, but I refused to do it.
date: 2025-07-30 draft: false title: Imperfect combination of Claude Code and Silicon Flow lang: en
Judging from the above test results, it seems that the model of the Kimi platform is really usable. If I used Silicon Flow, I would choose deepseek-ai/Deepseek-V3. Compared with its brother who didn’t do anything but claimed to have done everything, it was cheaper.