Alibaba's Qwen family is one of the strongest open-weight model lineups anywhere — and one of the hardest to actually pay for from outside China. NovaToken gives you a single OpenAI-compatible endpoint for Qwen (plus DeepSeek, GLM and Kimi), billed in USD on any card.
base_url to https://api.novatoken.dev/v1. Streaming, tool calling and JSON mode all work out of the box.
Qwen models punch far above their price: the 7B-class instruction model handles chat, classification and tool calling reliably, with roughly a two-second first token and no thinking pause. Developers use them for chatbots, data extraction pipelines, translation, and as cheap fallback models behind a bigger frontier model.
The catch is access. The usual routes — Alibaba Cloud international, or Chinese inference platforms — mean Chinese accounts, Alipay/WeChat top-ups, or enterprise verification. For a solo developer with a Visa card, that's a dead end. NovaToken exists to fix exactly that:
| Model | Best for | Price / 1M tokens |
|---|---|---|
| Qwen2.5-7B-Instruct | Fast chat, classification, tool calling | $0.10 in / $0.30 out |
| DeepSeek-V4-Flash | Reasoning, long documents (1M context) | $0.30 in / $1.20 out |
| GLM-5.3-Flash | Bilingual tasks, agents | $0.15 in / $0.50 out |
| Kimi-K2.7-Code | Code, agentic workflows (262K context) | $1.20 in / $5.00 out |
Fifty million input tokens on Qwen2.5-7B-Instruct for five dollars — enough to prototype a full product. No subscription, no expiry, top up again whenever you need.
Buy $5 of credits Create free account
Currently Qwen2.5-7B-Instruct (fast, no thinking mode). Larger Qwen models are added as demand grows — email [email protected] if you need a specific one.
Yes. Chat completions, streaming (SSE), function/tool calling and JSON mode are supported in every language with an OpenAI-compatible client.
Qwen2.5-7B-Instruct answers with roughly a 2-second first token. The reasoning models (DeepSeek, GLM, Kimi) pause a few seconds while thinking; with streaming the first bytes still arrive fast.
No. Billing metadata only, purged on a rolling basis. Request bodies are not retained.
Major credit and debit cards worldwide, charged in USD. Prepaid packs of $5 / $20 / $50.