backend

How to Access DeepSeek V4 Pro for Free: A Step-by-Step Guide Using ZenMux and Local Tools

By 神奇小汤圆 · original author ·Jun 14, 2026

Read original on juejin.cn ↗ Google Translate ↗ Backup translation

Why it matters

The author, known as '神奇小汤圆' (Magic Little Tangyuan), claims to have found a way to use DeepSeek V4 Pro for free via ZenMux. For English-speaking developers, this is relevant because it demonstrates a method to access a capable LLM without cost, using tools like cc-switch that can also bridge to Claude Code. It highlights a growing ecosystem of API aggregation platforms and local forwarding tools that enable flexible, low-cost experimentation with multiple models.

Summary

The article provides a practical walkthrough for developers to access DeepSeek V4 Pro's free tier without paying. It centers on ZenMux, a platform that acts as a gateway to the free models, and shows how to create an API key with model restrictions to ensure only free models are used. The guide covers two common local tool setups: Cherry Studio for quick testing and cc-switch for more flexible API forwarding, including integration with Claude Code.

The process involves registering on ZenMux, generating a restricted API key that limits access to specific free models like deepseek-v4-pro-free and deepseek-v4-flash-free, and then configuring either Cherry Studio or cc-switch with the key and appropriate API endpoints. The author notes that the free models have limitations such as token caps, potential slowdowns during peak hours, and reduced context length compared to paid versions, but considers the setup highly cost-effective for personal use and development testing.

The article also includes practical tips: using the flash-free model for simple tasks and the pro-free model for complex reasoning, avoiding public exposure of the API key, and segmenting long texts for better performance.

Key takeaways

— ZenMux provides free access to DeepSeek V4 Pro and other models through a restricted API key system.

— The free models include deepseek-v4-pro-free (stronger reasoning) and deepseek-v4-flash-free (faster, for simple tasks).

— cc-switch is a local API forwarding tool that can map models and integrate with tools like Claude Code.

— Cherry Studio can be used as an alternative to cc-switch for testing the API key.

— The API endpoint for cc-switch configuration is https://zenmux.ai/api/anthropic.

— Free models have token limits and may experience slowdowns during peak hours.

— API keys should never be exposed in public repositories to prevent abuse.

— Model restriction must be set when creating the API key to ensure only free models are called.

Our take

The article's claim of 'DeepSeek V4 Pro' is somewhat misleading — the model itself responds that it is not DeepSeek-v4, suggesting ZenMux might be routing to a different underlying model.

Using a third-party aggregator like ZenMux introduces a dependency: if ZenMux changes its free tier or goes offline, the setup breaks.

The guide assumes familiarity with Chinese developer tools (cc-switch, Cherry Studio) which may not be well-known to English-speaking developers.

The integration with Claude Code via cc-switch is a clever workaround to use a free model with a tool that typically requires paid API access.

The token limit and peak-hour throttling are significant constraints for production use, making this more suitable for prototyping or light personal use.

The article does not discuss rate limits or concurrent request restrictions, which could be a hidden limitation.

Concepts & terms

ZenMux

A platform that aggregates multiple AI model APIs, offering both free and paid tiers. It allows users to generate API keys with model restrictions to control which models can be accessed.

cc-switch

A local API forwarding and switching tool that enables developers to route API calls from local applications (like Claude Code) to different backend providers. It supports model mapping and custom configurations.

Cherry Studio

A desktop application for interacting with various large language models. It supports custom API providers and is often used for testing and managing multiple AI models from a single interface.

Model Restriction

A feature in API key management that limits which models a key can access. By specifying allowed model names, users can prevent accidental use of paid models and ensure only free or approved models are called.

API Forwarding

The practice of routing API requests through an intermediary tool (like cc-switch) that can modify the request, map model names, or change endpoints. This allows local tools to work with different API providers without changing their configuration.

Read the source: juejin.cn ↗ Google Translate ↗ Backup ↗