Free API keys for GPT and DeepSeek, but leave your business plan at home

What it does

This repository is essentially a brochure for a hosted API proxy that forwards requests to OpenAI, DeepSeek, Anthropic, Google, and other LLM providers through a single OpenAI-compatible endpoint. It hands out free API keys with tight daily limits, while funding itself through a paid tier that bills per token and drops the rate caps. The main draw for mainland Chinese users is domestic routing that promises to work without a VPN.

The interesting bit

There is no visible proxy code to self-host here—just documentation, status badges, and a long pricing table. The free tier is aggressively restricted: 200 requests per day per IP and key, with flagship models like the GPT-5 series limited to five daily calls. The authors also openly admit that free-tier GPT-5 models have “weaker reasoning ability” than their paid counterparts.

Key highlights

• Free access to GPT, DeepSeek, Claude, Gemini, Grok, and others via a unified OpenAI-style API.
• Domestic China endpoint (api.chatanywhere.tech) advertised as lower-latency than tunneling to official APIs.
• Hard rate limits on free keys: 200 requests/day per IP, with premium models capped at 5–30 requests.
• Paid tier claims lower-than-official per-token pricing and no expiration on prepaid balances.
• Privacy policy claims the proxy itself does not log prompts or caller identity, though it notes OpenAI retains data for 30 days.

Caveats

• Free keys are explicitly restricted to personal, educational, and non-profit research use; commercial use or model training will get keys banned.
• The maintainers state the system is “only for internal evaluation and testing,” and any commercial or public-facing use is at your own risk.
• Rate limits are enforced per IP across all keys, so sharing keys or rotating them from the same network does not help.

Verdict

Worth a look for individual Chinese developers running small personal experiments who need quick API access without a VPN. Skip it if you need a production-grade, self-hosted, or commercially licensed solution.