If you want to test an LLM API with Claude Code, Cursor, Codex CLI, Gemini CLI, or a desktop client such as Cherry Studio, the difficult part is often not the SDK call. It is finding a gateway that clearly documents its models, protocols, limits, and current trial credits.
I manually checked seven independent community LLM API gateways in August 2026. I visited their registration pages, model catalogs, public announcements, and integration docs.
Disclosure: These are independent third-party services, not official OpenAI, Anthropic, or Google endpoints. The detailed comparison linked below contains referral links that may grant platform credits to both the new user and the referrer. Offers can change at any time.
Quick selection guide
| What you need | Gateways to check first |
|---|---|
| A visible new-user credit system | Xiaoen AI |
| Claude Code, Codex, Gemini CLI, and other coding tools | Agent Router or DoCode |
| Cherry Studio and CC Switch integration | Fuka API |
| Free models plus daily check-ins | Liminality |
| A public model catalog and billing style | Kemeng API |
| A small Claude-focused routing option | GoRouter |
What stood out
1. Xiaoen AI
Its public pages showed one of the clearest combinations of registration credits, referral credits, and daily check-ins. The exact numbers were not fully consistent across every page, so treat them as current promotional information rather than a permanent promise.
Best for: running a first set of API experiments with trial credits.
2. Agent Router
This is positioned more like a multi-protocol developer gateway. Its public documentation includes endpoints compatible with:
- OpenAI Chat Completions
- OpenAI Responses
- Anthropic Messages
- Gemini-style model APIs
- Embeddings, reranking, image, and audio workflows
The docs also mention Claude Code, Codex, Gemini CLI, Qwen Code, Cursor, Cline, and Roo Code.
Best for: developers who switch between several AI coding tools.
3. Fuka API
The public site lists model families from OpenAI, Anthropic, Google, DeepSeek, Qwen, and others. It also provides client-oriented setup information for Cherry Studio and CC Switch.
Best for: users who prefer configuring a desktop client instead of writing integration code.
4. DoCode
DoCode organizes much of its documentation around actual developer workflows. It covers Claude Code, Codex CLI, Gemini CLI, Cursor, Cline, Aider, Cherry Studio, and Chatbox, with OpenAI-compatible, Responses, Claude-native, and Gemini-native options.
Its promotional credit text was not synchronized across every public page when I checked it, so verify the amount after registration.
Best for: developers who want tool-specific setup guides.
5. Liminality
Liminality uses an internal credit called an “orange.” Its public catalog showed 39 models at the time of checking, including several models priced at zero oranges, plus a daily check-in system.
Its terms also state a per-key request limit and no commercial SLA.
Best for: personal learning, occasional prompts, and lightweight model comparison — not production traffic.
6. Kemeng API
This gateway exposes a public model catalog and an OpenAI-compatible base URL. The catalog displayed 29 models when checked, with per-request and usage-based billing styles plus some latency and success-rate information.
A referral promotion visible during the audit had a specific August 2026 expiration date, so verify the current terms.
Best for: inspecting models and doing controlled API integration tests.
7. GoRouter
GoRouter presents a smaller unified API gateway with Chat and Responses-style interfaces and client setup links. Its public catalog was mainly focused on a few Claude routes, and the public registration page did not promise a fixed signup reward.
Best for: a simple Claude-routing experiment.
A safer evaluation workflow
For any third-party gateway:
- Create a dedicated test API key.
- Set a low spending or usage limit if available.
- Copy the exact base URL and model identifier from its dashboard.
- Run the same short prompt through each service.
- Compare latency, output quality, tool calling, error behavior, and billing records.
- Keep official APIs or multiple replaceable providers for important workloads.
Never send company source code, customer data, personal information, long-lived secrets, or other sensitive material through an untrusted gateway. Free credits are useful for learning and integration testing; they are not evidence of production reliability.
The full comparison includes screenshots/claims checked against public pages, current credit notes, tool compatibility, and registration links:
https://haha0510.github.io/llm-api-public-stations/
The detailed page is currently in Chinese, but its comparison table and integration notes are structured for quick translation.
Top comments (0)