CatRouter Review: Pricing & Comparison
Mainland-direct aggregator at ¥1 = $1 of credit, with group multipliers published per group (claims 500+ models; 162 actually listed)
Last verified: 2026-10-15 · Visit official site →
A mainland-direct aggregator built on ¥1 = $1
CatRouter (catrouter.net) positions itself as an “enterprise-grade AI management platform”, leading with 500+ models, 100,000+ developers served, 99.9% availability and a 320ms average response. It describes itself as having been “merged from several large model aggregation platforms, rarely promoted publicly, mainly serving existing users” — which explains why it barely appears in Chinese developer discussions yet is still running steadily.
Pricing is the blunt selling point: ¥1 RMB buys $1 of model credit, from a ¥1 minimum, billed through per-group multipliers. Its documentation spells out the arithmetic — computing at a USD rate of 6.8, ¥1 buys $1 of credit, an effective saving of about 85.3%. A multiplier of 1 works out to roughly 15% of official pricing, the discounted Codex group gets to about 12%, while domestic models (DeepSeek/GLM/Kimi) bill at official list price with no exchange-rate subsidy at multiplier 1.
Group multipliers (published on the pricing page, partial):
- Discounted Codex-only: 0.6
- deepseek-special: 0.66
- flux image generation: 0.7
- codex independent / Codex-only: 0.8
- kimi-k2.7-code special: 0.8
Image examples: gpt-image-2-1k ¥0.029/image, gpt-image-2-d ¥0.07/image (2K), gpt-image-2-c ¥0.12/imageWorth noting: the trade-off between cheap and stable is stated openly — “the lower the multiplier, the less stable; choose according to need” — and the support site concedes that official OpenAI/Claude channels slow down under load, and that discounted groups can get slower at peak.
Coverage: “500+” claimed, 162 actually listed
The homepage says “500+ Access Model”, but the model plaza filter on the pricing page shows 162 models, broken down as OpenAI 38, Google 28, Alibaba Bailian 25, Anthropic 20, DeepSeek 9, xAI 6. By endpoint type: openai 130, openai-response 27, anthropic 25, gemini 24, image-generation 11 — so it is not merely an OpenAI-compatible wrapper; Claude and Gemini native protocols are served directly.
That gap between the marketing number and the actual catalogue is worth keeping in mind: the pricing page, not the homepage, is the real list of what you can call. The contact page also lists a phone number that looks like a placeholder, and no company name or ICP filing is published, so the legal entity cannot be independently verified.
Who it’s for
If you want one key covering OpenAI, Claude and Gemini native protocols plus image generation at 10–20% of official pricing, CatRouter belongs on your shortlist. Accept two realities first: discounted groups really are less stable (their own words), and the operating entity is opaque. That makes it a poor place to park a critical workload long-term — small top-ups with a backup provider is the more sensible pattern.
Information verified 2026-10-15. Pricing, model counts and group multipliers reflect what catrouter.net and its pricing page displayed on that date; relay groups and multipliers change frequently.
Related reviews
- PoloAPI: balances stability and multi-model coverage, focused on usage statistics and multi-project call auditing
- 4SAPI / Xinglian 4SAPI: an enterprise-grade model aggregation hub, focused on low latency and corporate invoicing
- AiHubMix: developer-integration-friendly, permanently free test tier, Prompt Caching support, thorough documentation
- SiliconFlow: domestic open-source LLM cloud service, 100+ models including DeepSeek/Qwen/GLM, self-built inference engine + domestic chip support
Quick facts
| Pricing model | Top up ¥1 RMB = $1 of model credit, from ¥1; billing runs on per-group multipliers — the default multiplier of 1 works out to roughly 15% of official pricing, the discounted Codex group reaches about 12%, and domestic models (DeepSeek/GLM/Kimi) bill at official list price at multiplier 1; login bonus credit and a daily check-in bonus |
|---|---|
| Model coverage | The homepage claims 500+ models; the pricing page's model plaza actually lists 162 (OpenAI 38 / Google 28 / Alibaba Bailian 25 / Anthropic 20 / DeepSeek 9 / xAI 6), across endpoint types openai 130, openai-response 27, anthropic 25, gemini 24, image-generation 11 |
| Latency / SLA | Claims 99.9% availability and a 320ms average response, with no third-party verification; its own support site concedes that slow-downs in official OpenAI/Claude upstreams affect them too and that discounted groups can get slower at peak |
| Mainland direct connect | Mainland direct-connect (claims return-path optimisation for all three major carriers, no proxy needed) |
| Best for | Developers / Enterprise |
| Referral program | No public affiliate/rebate programme found; the operator says it no longer runs a user community group for new signups. |
Pros
- Unusually granular price transparency: the site publishes the multiplier for each routing group (discounted Codex-only 0.6, deepseek-special 0.66, flux image 0.7, codex-only 0.8) and warns outright that the cheaper the multiplier, the less stable the group
- One key spans OpenAI, Anthropic and Gemini native protocols plus image generation endpoints (openai 130 / openai-response 27 / anthropic 25 / gemini 24 / image-generation 11), wider protocol coverage than relays that only wrap everything in an OpenAI-compatible layer
- A separate support site (service.catrouter.net) publishes service hours and a ticket channel, and self-service invoicing is available
Cons
- The homepage advertises 500+ models while its own model plaza lists 162 — a clear gap between marketing and what is actually purchasable
- Exposed to the same upstream dynamics as everyone else: it admits official OpenAI/Claude channels slow down under load, and discounted groups can be slower at peak
- The operating entity is opaque — no company name on the site, a phone number that looks like a placeholder, and no visible ICP filing, so the legal entity cannot be independently verified
Compare more AI API relays
See the full comparison board — filter by price tier, model coverage, and mainland direct-connect status.
Back to the comparison board →