One key for the newest flagships from all four of China's frontier labs — Qwen 3.8 Max, Kimi K3, DeepSeek V4 and GLM 5.2 — the official, full-quality models, not quantized community mirrors, at up to 48% below the vendors' international rates — cached tokens included. Native to Claude Code, Cline and Cursor in two lines of config.
# two lines. that's the whole migration. export ANTHROPIC_BASE_URL="https://api.gloritoken.ai" export ANTHROPIC_AUTH_TOKEN="sk-your-key" $ claude ✔ Connected · model: qwen3.8-max > refactor this module and add unit tests…
Qwen 3.8 · Kimi K3 · DeepSeek V4 · GLM 5.2 — the newest flagship from each of China's four frontier labs, first-hand from the makers' official APIs, full-quality builds. Priced up to 48% below the vendors' international rates — Qwen 3.8 Max at 48% below, Kimi K3 freshly cut to 42% below, and cached input discounted at the same ratio.
* Pay-as-you-go, USD per 1M tokens. Alibaba-served models are 12–48% below the vendors' international list rates, and cached input is discounted at the same ratio — the cache maths works exactly as it does going direct. Two official-direct models (deepseek-v4-pro, kimi-k2.7-code-highspeed) are at vendor parity; we earn nothing on them. Full breakdown, including who should not use us. Models run on the makers' official APIs (mainland-China endpoints) via our Singapore gateway; we don't store prompt contents and never train on your data. We do not relay Claude, GPT or Gemini.
No subscription, no minimum commitment — you pay only for the tokens you use, at up to 48% below the model makers' international rates.
* Prepaid credit, billed in USD. Consumed usage is non-refundable; any unused balance can be refunded on request within 30 days — see Terms. Questions about billing: hello@gloritoken.ai.
Anything that speaks the OpenAI or Anthropic API works out of the box — point it at gloritoken and you're running on China's frontier coding models.
* Tool names and logos belong to their respective owners; shown for compatibility only, no affiliation or endorsement implied.
Our gateway runs in Singapore; inference happens on the model makers' official APIs, which are served from mainland China — we state that plainly rather than hide it. We don't store your prompt contents and never train on your data; we keep only the usage metadata needed for billing.
OpenRouter routes you to whichever provider is cheapest — often a quantized, community-hosted copy that varies in quality and uptime. gloritoken serves the official, full-quality models first-hand, natively over the Anthropic protocol (real Claude Code support), with Alibaba-served models priced 12–48% below the vendors' international rates — and one key covers every lab.
Yes — natively. We expose a real Anthropic-protocol endpoint (/v1/messages), so Claude Code connects with two environment variables, no proxy hacks. Cline, Roo Code, Cursor and any OpenAI-compatible tool work the same way through /v1.
The API is live and self-serve: sign up, verify your email, and start with free credit — instantly, no waiting. Need a bigger trial or a team plan? Tell us what you're building.