Get one Claude Code API key for Claude, Kimi, GLM, DeepSeek, and StepFun
Claude Code talks the Anthropic Messages format. Point it at this Messages-compatible gateway, buy prepaid credits, and switch model routes when Claude is too expensive or overloaded.
Two ways to get Claude Code API access
Claude Code only needs a Messages-compatible endpoint and a key. The choice is about cost, model flexibility, and how fast you can start.
| Compatible gateway | First-party Anthropic | |
|---|---|---|
| Endpoint Claude Code needs | Messages compatible — works out of the box | Messages (first-party) |
| Models you can switch to | Claude, plus Kimi, Qwen, GLM, DeepSeek, MiMo via one env var | Claude family only |
| Cost control | Route to a 0.1x–2x model when Claude is overkill | Claude pricing only |
| Start time | Buy online, key in minutes | Account and billing setup |
| Credits | Prepaid CNY, no expiry | USD usage billing |
| Enterprise SLA | Not offered today | Available |
Who should buy which
Buy the gateway key if you
- Want to lower Claude Code's per-token cost by routing to cheaper models for routine work
- Want one key that runs Claude Code today and other models tomorrow
- Need to start now without a procurement cycle
- Like prepaid credits that don't expire
Buy direct from Anthropic if you
- Need a first-party SLA and signed data terms
- Must be on the newest Claude model the day it launches
- Require USD invoicing for procurement
- Have an enterprise agreement that mandates first-party access
Honest limitations
What the gateway route does and does not give you.
- Switching ANTHROPIC_MODEL to a non-Claude model changes behavior — agentic coding quality varies by model.
- This is a compatible gateway, not Anthropic; there is no first-party SLA.
- Billing is in CNY prepaid credits, not USD.
- We do not publish latency numbers we have not measured. Test your own Claude Code session before committing.
Real Claude Code costs on this gateway
Published per-model rates. Multipliers apply to a 0.4 CNY per 1M token base.
Coding quality and latency vary by model and task. We do not publish benchmark or latency claims we have not measured — run a real Claude Code session against the docs endpoint to decide.
Claude Code API — FAQ
Does Claude Code work with this API?
Yes. Claude Code uses the Anthropic Messages format. Set ANTHROPIC_BASE_URL to the gateway and ANTHROPIC_MODEL to claude-sonnet-4-6 or any supported model.
Can I run Claude Code on cheaper models?
Yes. Change ANTHROPIC_MODEL to a model like kimi-k2.6, glm-5.1, or a 0.1x route. Quality varies by model, so test on your own tasks.
How much does Claude Code cost here?
It depends on the model. Rates run from 0.04 CNY per 1M tokens (0.1x models) up the multiplier scale. The default Sonnet route and the full model list are on the pricing page.
Should I just use Anthropic directly?
If you need a first-party SLA or the newest model on launch day, use Anthropic. If you want instant access, multi-model routing, and prepaid credits, the gateway is the faster, cheaper path.
Get Claude Code API access in minutes
Buy online, set three environment variables, and message Telegram support if you want help choosing the first model route.