Skip to content
Public beta

OpenAI-compatible API for frontier open models

DeepSeek V4 · Kimi K2 · GLM

Up to 40% off upstream pricing · VietQR · prepaid

No credit card · billed per real token · works with the OpenAI SDK unchanged

curl https://api.clfaigateway.dev/v1/chat/completions \  -H "Authorization: Bearer sk-gw-..." \  -H "Content-Type: application/json" \  -d '{    "model": "kimi-k2.6",    "messages": [{"role": "user", "content": "Hello!"}],    "stream": true  }'

Models & pricing

Prices in USD per 1M tokens. Upstream model IDs are public — you know exactly what runs.

ModelCapabilitiesContextInputCached inputOutput
glm-5.240% off@cf/zai-org/glm-5.2toolsreasoningjson262K$1.40$0.84$0.26$0.156$4.40$2.64
kimi-k2.7-code40% off@cf/moonshotai/kimi-k2.7-codetoolsvisionreasoning262K$0.95$0.57$0.19$0.114$4.00$2.40
kimi-k2.640% off@cf/moonshotai/kimi-k2.6toolsvisionreasoning262K$0.95$0.57$0.16$0.096$4.00$2.40
glm-4.7-flash40% off@cf/zai-org/glm-4.7-flashtoolsreasoningfast131K$0.06$0.036$0.40$0.24
deepseek-v4-flash40% off@cf/deepseek-ai/deepseek-v4-flash-0731toolsreasoningfast1M$0.44$0.264$0.014$0.008$1.32$0.792
deepseek-v4-pro40% off@cf/deepseek-ai/deepseek-v4-pro-0813toolsreasoningjson1M$1.32$0.792$0.044$0.026$3.96$2.376
qwen3.8-27b40% off@cf/qwen/qwen3.8-27btoolsreasoningvision262K$0.45$0.27$0.45$0.27$3.20$1.92

Full cache hits are billed at 10% of the normal price. Prices as of 2026-08-18 — dashboard shows live rates.

All models & pricing →
HTTP/2 200content-type: text/event-streamx-gw-provider: cloudflare-workers-aix-gw-model: kimi-k2.6x-gw-cache: missx-ratelimit-limit: 60

Response headers from a real request, content omitted.

You know exactly what you buy

CLF AI Gateway runs on Cloudflare Workers AI — and says so on the tin. Every model's upstream ID is public, and every response tells you which model actually ran.

  • Upstream IDs public: every model page shows its @cf/… identifier
  • Every response carries x-gw-provider and x-gw-model headers
  • Performance numbers are published with their measurement conditions
  • System status is public — no login required
Gateway overhead p9513.6 msn = 6,033 requests · staging under sustained load · Aug 2026
Time to first token, median0.65 sglm-4.7-flash · streamed via edge · 400 requests, Aug 2026
Input price from, per 1M tokens$0.036glm-4.7-flash · standard rate
Public upstream IDs7 modelsLive status →

Up and running in three steps

01

Create an account

Sign up and get your sk-gw-… API key. No credit card, about 30 seconds.

sk-gw-****************
02

Top up credits

VietQR bank transfer auto-confirms in ~15 s. International cards work too.

topup.confirmed · ~15 s
03

Point your SDK at us

Change base_url, keep your OpenAI SDK. Streaming, tools and JSON mode included.

base_url = "https://api.clfaigateway.dev/v1"

Frequently asked questions

Open-weight frontier families: DeepSeek V4 (DeepSeek), Kimi (Moonshot AI), GLM (Z.ai) and Qwen (Alibaba). DeepSeek V4 brings a 1M-token context window, Kimi is strong at coding and agentic tool use, GLM at reasoning, and Qwen and Kimi accept image input; all seven models support tool calls and streaming.

See all models →

You can — the model IDs are public. The gateway adds prepaid USD billing, VietQR top-ups, per-request cost logs, automatic caching discounts and one OpenAI-compatible endpoint, at 13.6 ms p95 measured overhead.

You top up prepaid credits; each request deducts its exact token cost at the model's current price. At zero balance new requests fail fast with a clear error — no subscriptions, no surprise invoices.

Billing docs →

Tokens already generated and delivered are charged. A request that failed before producing any output costs $0 — that is the zero-completion insurance.

Billing docs →

No. Logs are metadata-only: token counts, cost, latency, model. Prompt and response content is processed in memory and discarded — the privacy policy is written to match.

Privacy policy →

Each organization gets requests-per-minute, tokens-per-minute and concurrency limits, and every response reports them in x-ratelimit-* headers. Need more headroom? Email us.

Used credits are not refundable. Unused credits can be refunded case-by-case within 14 days of purchase, and billing errors on our side are always made whole.

Terms →

VietQR bank transfer, auto-confirmed in about 15 seconds, or international cards through Dodo Payments as Merchant of Record. Purchased credits don't expire.

Get your API key in 30 seconds

No credit card · prepaid · works with the OpenAI SDK