OpenRouter is a catalogue — one key for dozens of providers. NovaToken is a specialty shop: four Chinese open-weight models, list-price token billing, and a router that picks the right one per request. Which is better depends entirely on what your traffic looks like.
| OpenRouter | NovaToken | |
|---|---|---|
| Model coverage | Broad: Western labs + many open-weight providers | Focused: DeepSeek-V4-Flash, Qwen2.5-7B, GLM-5.3-Flash, Kimi-K2.7-Code |
| API shape | OpenAI-compatible | OpenAI-compatible |
| Payment | Card | Card, USD, no Chinese phone or Alipay |
| Token pricing | Provider list price + platform fee structure | Matches the official list price for the models we serve |
| Minimum top-up | Small | $5, credits never expire |
| Model selection | You choose; provider fallbacks available | You choose — or send auto and each request is routed by difficulty |
| Billing transparency | Per-request cost visible in dashboard | Per-request cost plus the real model name in your usage log, broken down per model |
| Support | English, community-driven | English and Chinese, direct email |
| Best for | Multi-vendor apps, experiments across labs | Apps built on Chinese open-weight models |
Both services change pricing and catalogue regularly — verify current numbers on each provider's site before you commit a budget.
Two fields change. Model ids are the only thing to watch — they differ from OpenRouter's naming.
If you would rather not name a model at all, send "model": "auto". Short chat goes to Qwen, everyday answers to GLM, long or tool-heavy requests to DeepSeek — and you are billed at whichever model actually answered, at that model's normal rate.
Work out your monthly mix first: how many input tokens, how many output tokens, on which model. Then multiply by the list price. On DeepSeek-V4-Flash that is $0.30 / 1M input and $1.20 / 1M output, with cache-hit input at $0.006 — the cache discount is usually the single biggest line item for agents and coding tools, because they resend the same context every turn.
A workload of 10M input + 2M output on DeepSeek-V4-Flash costs about $5.40 a month at list price. Compare that with what you actually paid last month on any gateway — that difference, not marketing copy, is the real answer.
Yes, and many teams do: NovaToken for bulk Chinese-model traffic, another gateway for the occasional GPT or Claude call. Both speak the OpenAI API, so the client code differs only in base_url and key.
No. Ours use the upstream names, e.g. deepseek-ai/DeepSeek-V4-Flash, Qwen/Qwen2.5-7B-Instruct, zai-org/GLM-5.3-Flash, moonshotai/Kimi-K2.7-Code. Copy them from the pricing table.
auto?No. You pay the routed model's own per-token price. The routing decision is free, and your usage log records which model ran.
Requests fall back to the New API backend automatically. If something upstream is degraded, email support and we will tell you what is happening rather than leaving you to guess.
Same OpenAI API, four Chinese models, card payment in USD, no routing fee.
Buy $5 of credits