OpenAI compatible API · Attested · Public status

Chinese AI Models: Labs, Routes & Where Prompts Go

Chinese-lab models in the catalog, the providers that serve each one, and how to run those weights on a US-operated route instead of the vendor endpoint.

Verify gateway
Onebase URL to migrate
100sof models and routes
0prompt or output logs. Always.
Chinese labs, and the operators who serve their weights

A Chinese model does not mean China-bound traffic. Those are two different facts, and you choose the second one.

237 models in the TrustedRouter catalog were built by one of the 12 labs recorded as Chinese: Alibaba (Qwen), Z.ai (GLM), DeepSeek, ByteDance (Seed, Doubao), MiniMax, Moonshot AI (Kimi), Tencent (Hunyuan), Xiaomi (MiMo), Ant Group (inclusionAI: Ling, Ring), Baidu (ERNIE), Kuaishou (Kwaipilot KAT), StepFun. Every model below is grouped under the lab that built it, with its source.

Where your prompt goes is a separate decision, made per request. Most of these are open-weights models, so other companies serve them: 43 of the 95 providers in the catalog are US-headquartered, and the "US-operated routes" column below names the ones serving each model. Route to one of those and your prompt reaches that provider, not the lab. Route to the vendor's own API and it reaches a China-headquartered operator — 5 of the providers here.

The lab a model came from and the operator of each of its routes are recorded in separate catalog tables, and neither is derived from the other. Read both before deciding.

US-hosted Chinese models Browse every route

us_hosted_chinese_model.pyPython
from openai import OpenAI

client = OpenAI(
    base_url="https://api.trustedrouter.com/v1",
    api_key="sk-tr-...",  # your TrustedRouter key
)

# Chinese-built weights, US-operated endpoint:
# jurisdiction narrows routing to providers whose
# recorded operator country is the United States,
# and fails closed if none qualify.
reply = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro",
    messages=[{"role": "user",
               "content": "Refactor this handler."}],
    extra_body={"provider": {"jurisdiction": "us"}},
)
The lab's own APIA US-operated route for the same weights
Who built the modelThe Chinese labThe same Chinese lab. Routing does not change this.
Who operates the endpointA company whose operating home is China — named in its own policy where one is published, otherwise the group’s own filingA company headquartered in the United States, named in its own terms or filing
Where the request is sentTo the lab's operatorTo that US provider. The request is not sent on to the originating lab.
How you select itCall the vendor route directlySet provider.jurisdiction to us, or pin operators with provider.only
Retention postureWhatever that vendor's policy saysA separate field per provider. Jurisdiction and retention are independent; the table below shows both.
What neither one provesWhich datacentre answeredWhich datacentre answered. Operator country is a legal home, not a machine location.

Where the brand and the operator come apart

read the row, not the logo

The provider table below lists 5 providers operated from China. Some familiar Chinese AI brands are not among them, and that is the point of recording operators rather than brands.

Live catalog, as of September 2026

Two separate facts: who operates the endpoint, and who built the model.

Counts below are read from the routes configured in TrustedRouter right now. Operator country is the legal home of the company running the endpoint a request reaches, taken from that company's own terms, privacy policy, or regulatory filing. It is not a claim about which datacentre serves the request.

5providers operated from China
12labs with models here
237models from those labs
95providers in the catalog
529callable catalog models
29providers whose operator we could not establish

Providers operated from China

5 of 95

Send a request to one of these and the operator is a company headquartered in China — head office, not place of registration. Privacy tier is a separate column: it is the highest bar we have recorded from that provider's published policy.

Provider Operator country Privacy tier Recorded claims Models served Policy source
Alibaba Cloud Model Studioalibaba China Standard no verified privacy claim 77 policy
DeepSeekdeepseek China Standard no verified privacy claim 6 policy
StepFunstepfun China Standard no verified privacy claim 3 policy
Kling AIkling China Standard no verified privacy claim 2 policy
Xiaomi MiMoxiaomi China Standard no verified privacy claim 2 policy

Every provider's full policy note and route list is on its own page. 29 providers in the catalog publish no operator TrustedRouter could pin down, so they appear in no jurisdiction list on this site and no request preference treats them as matching one. Review every provider.

Models built by Chinese labs

237 models

Grouped by the lab that built the weights, with the source TrustedRouter read that from. The operator columns show who runs the endpoints for each route, which is a different question from the lab's country and frequently a different answer.

Alibaba (Qwen) 112 models

Alibaba Cloud states it provides the Qwen model series to the open-source community; Alibaba Group Holding Limited gives its principal executive offices as Hangzhou, China (https://www.alibabagroup.com/en-US/faqs-corporate-information). Source.

Model Context Routes Operator jurisdictions US-operated routes
qwen/qwen3.8-27b
Qwen: Qwen3.8 27B
262,144 19 +3 more on the model page
qwen/qwen3.5-397b-a17b
Qwen: Qwen3.5 397B A17B
262,144 17 +3 more on the model page
qwen/qwen3.6-35b-a3b
Qwen: Qwen3.6 35B A3B
262,144 17 +1 more on the model page
qwen/qwen3.6-27b
Qwen: Qwen3.6 27B
262,144 14 +1 more on the model page
qwen/qwen3.5-122b-a10b
Qwen: Qwen3.5-122B-A10B
262,144 12
qwen/qwen3-next-80b-a3b-instruct
Qwen: Qwen3 Next 80B A3B Instruct
262,144 10

106 further Alibaba (Qwen) models are in the catalog, ranked below these by route count. Browse every model.

Z.ai (GLM) 29 models

The GLM family comes from the Beijing-headquartered lab that renamed itself Z.ai in 2025, formerly Beijing Zhipu Huazhang Technology Co., Ltd. The API TrustedRouter routes to is operated by a Singapore entity (see the zai provider row), which is a separate fact. Source.

Model Context Routes Operator jurisdictions US-operated routes
z-ai/glm-5.2
Z.ai: GLM 5.2
1,048,576 44 +12 more on the model page
z-ai/glm-5.3
Z.ai: GLM 5.3
1,048,576 42 +10 more on the model page
z-ai/glm-5.3-flash
Z.ai: GLM 5.3 Flash
1,048,576 42 +12 more on the model page
z-ai/glm-5.1
Z.ai: GLM 5.1
202,752 20 +1 more on the model page
z-ai/glm-4.7
Z.ai: GLM 4.7
202,752 11
z-ai/glm-5
Z.ai: GLM 5
202,752 10

23 further Z.ai (GLM) models are in the catalog, ranked below these by route count. Browse every model.

DeepSeek 26 models

DeepSeek's privacy policy names Hangzhou DeepSeek Artificial Intelligence Co., Ltd., registered in China, as data controller. Source.

Model Context Routes Operator jurisdictions US-operated routes
deepseek/deepseek-v4-flash-0731
DeepSeek: DeepSeek V4 Flash 0731
1,048,576 33 +7 more on the model page
deepseek/deepseek-v4-flash
DeepSeek: DeepSeek V4 Flash 0423
1,048,576 27 +6 more on the model page
deepseek/deepseek-v4-pro
DeepSeek: DeepSeek V4 Pro 0423
1,048,576 26 +5 more on the model page
deepseek/deepseek-v3.2
DeepSeek: DeepSeek V3.2
163,840 17 +2 more on the model page
deepseek/deepseek-v4.1-flash
DeepSeek: DeepSeek V4.1 Flash
1,048,576 17 +1 more on the model page
deepseek/deepseek-v4-flash-vision-exp
DeepSeek: DeepSeek V4 Flash Vision Exp
1,048,576 12

20 further DeepSeek models are in the catalog, ranked below these by route count. Browse every model.

ByteDance (Seed, Doubao) 18 models

ByteDance is headquartered in Haidian, Beijing, China, while its associated entity ByteDance Ltd is incorporated in the Cayman Islands. ByteDance's own site lists offices in roughly 120 cities without naming a headquarters. Source.

Model Context Routes Operator jurisdictions US-operated routes
bytedance/seed-1.8
ByteDance/Seed-1.8
256,000 2
bytedance/seed-2.0-code
ByteDance/Seed-2.0-code
256,000 2
bytedance/seed-2.0-mini
ByteDance/Seed-2.0-mini
256,000 2
bytedance/seed-2.0-pro
ByteDance/Seed-2.0-pro
256,000 2
bytedance/ui-tars-1.5-7b
ByteDance: UI-TARS 7B
128,000 2
bytedance/doubao-seed-1.6-251015
Doubao Seed 1.6
262,144 1

12 further ByteDance (Seed, Doubao) models are in the catalog, ranked below these by route count. Browse every model.

MiniMax 16 models

MiniMax Group Inc. is based in Shanghai, China and listed in Hong Kong (SEHK: 100). Source.

Model Context Routes Operator jurisdictions US-operated routes
minimax/minimax-m3
MiniMax: MiniMax M3
524,288 21 +5 more on the model page
minimax/minimax-m2.5
MiniMax: MiniMax M2.5
196,608 13
minimax/minimax-m2.7
MiniMax: MiniMax M2.7
196,608 11
minimax/minimax-m2
MiniMax: MiniMax M2
204,800 4
minimax/minimax-m2.5-highspeed
MiniMax M2.5 Highspeed
204,800 4
minimax/minimax-m2.1
MiniMax: MiniMax M2.1
204,800 3

10 further MiniMax models are in the catalog, ranked below these by route count. Browse every model.

Moonshot AI (Kimi) 16 models

Beijing Moonshot AI Technology Co., Ltd. gives its address as Haidian District, Beijing, China. Source.

Model Context Routes Operator jurisdictions US-operated routes
moonshotai/kimi-k3
MoonshotAI: Kimi K3
1,048,576 39 +9 more on the model page
moonshotai/kimi-k2.6
MoonshotAI: Kimi K2.6
262,144 33 +10 more on the model page
moonshotai/kimi-k2.7-code
MoonshotAI: Kimi K2.7 Code
262,144 19 +3 more on the model page
moonshotai/kimi-k2.5
MoonshotAI: Kimi K2.5
262,144 12
moonshotai/kimi-k2-thinking
MoonshotAI: Kimi K2 Thinking
262,144 4
moonshotai/kimi-k2.7-code-highspeed
moonshotai/Kimi: K2.7-Code-Highspeed
262,144 3

10 further Moonshot AI (Kimi) models are in the catalog, ranked below these by route count. Browse every model.

Tencent (Hunyuan) 4 models

Tencent Holdings Limited's 2018 annual report gives its group head office as Tencent Binhai Towers, Nanshan District, Shenzhen, PRC, with a principal place of business in Hong Kong. Source.

Model Context Routes Operator jurisdictions US-operated routes
tencent/hy3
Tencent: Hy3
262,144 8
tencent/hunyuan-a13b-instruct
Tencent: Hunyuan A13B Instruct
131,072 2 none configured
tencent/hy4-preview
tencent/hy4-preview
262,144 2
tencent/hy3-preview
Tencent: Hy3 preview
262,144 1

Xiaomi (MiMo) 4 models

Xiaomi Corporation's 2025 annual report gives its head office and principal place of business as Xiaomi Campus, Haidian District, Beijing, PRC, with a Cayman Islands registered office. Source.

Model Context Routes Operator jurisdictions US-operated routes
xiaomi/mimo-v2.5
Xiaomi: MiMo-V2.5
1,050,000 8
xiaomi/mimo-v2.5-pro
Xiaomi: MiMo-V2.5-Pro
1,050,000 8
xiaomimimo/mimo-v2.5
Xiaomi MiMo V2.5
1,048,576 2
xiaomimimo/mimo-v2.5-pro
Xiaomi MiMo V2.5 Pro
1,048,576 2

Ant Group (inclusionAI: Ling, Ring) 3 models

The inclusionAI organisation describes itself as the home of Ant Group's AGI initiative, and Ant Group's own offices page gives its principal business office as Z Space, No. 556 Xixi Road, Hangzhou, China (https://www.antgroup.com/en/about/our-offices). Source.

Model Context Routes Operator jurisdictions US-operated routes
inclusionai/ling-3.0-flash
inclusionAI: Ling 3.0 Flash
131,072 4
inclusionai/ling-3.0-flash-fin
inclusionAI: Ling 3.0 Flash Fin
262,144 2
inclusionai/ling-3.0-flash-vl
inclusionAI: Ling 3.0 Flash VL
131,072 2

Baidu (ERNIE) 3 models

Baidu, Inc.'s Form 20-F gives principal executive offices at Baidu Campus, No. 10 Shangdi 10th Street, Haidian District, Beijing, PRC. Source.

Model Context Routes Operator jurisdictions US-operated routes
baidu/cobuddy
CoBuddy
131,072 2
baidu/ernie-4.5-21B-a3b
ERNIE 4.5 21B A3B
120,000 2
baidu/ernie-4.5-vl-424b-a47b
Baidu: ERNIE 4.5 VL 424B A47B
123,000 2

Kuaishou (Kwaipilot KAT) 3 models

The Kwaipilot organisation describes itself as the AI team from Kuaishou, whose head office is in Haidian District, Beijing, PRC per its HKEX annual report (https://www1.hkexnews.hk/listedco/listconews/sehk/2022/0419/2022041900053.pdf). Source.

Model Context Routes Operator jurisdictions US-operated routes
kwaipilot/kat-coder-air-v2.5
KAT Coder Air V2.5
262,144 2
kwaipilot/kat-coder-pro-v2
KAT Coder Pro V2
262,144 2
kwaipilot/kat-coder-pro-v2.5
KAT Coder Pro V2.5
262,144 2

StepFun 3 models

StepFun's terms name Shanghai StepFun Intelligent Technology Co., Ltd. as the company providing its models and services. Source.

Model Context Routes Operator jurisdictions US-operated routes
stepfun/step-3.7-flash
StepFun: Step 3.7 Flash
262,144 3
stepfun/step-3.5-flash
StepFun: Step 3.5 Flash
262,144 1 none configured
stepfun/step-3.5-flash-2603
Step 3.5 Flash 2603
262,144 1 none configured

Model origin is keyed by the vendor prefix of a model id, and a prefix earns a row once at least three of its models are in the catalog. 42 callable catalog models sit under a prefix below that line and carry no recorded origin, so they are on no page here rather than assigned a country by guesswork.

Browse every modelReview provider policiesAI models from the United StatesAI models from the European Union

Quality

Judge the weights on the weights.

Benchmarks, price, and measured latency per route are on each model page and the leaderboard. Origin country belongs in a procurement review, not in a quality argument.

Control

Two preferences do the work.

provider.jurisdiction set to us restricts routing to US-operated providers. provider.only pins an exact allowlist. Both fail closed rather than falling back to something you did not approve.

Limits

What this page does not claim.

No claim that a US operator stores nothing, that any route is certified, or that operator country tells you where the hardware sits. Retention posture is the ZDR and confidential compute question, tracked separately.

Live catalog evidence

Current routes, prices, privacy, and measured performance.

Catalog facts come from the routes currently configured in TrustedRouter. Performance uses the same cached metadata snapshot as the public leaderboard. Prompts and outputs are not part of these measurements.

556public models
95providers
1767configured routes
309ZDR routes
35provider E2EE routes
7467recent availability samples
Model Providers Context Input Output Privacy Measured route
DeepSeek: DeepSeek V4 Flash 0731deepseek/deepseek-v4-flash-0731
+21
33 routes
1,048,576 $0.047475/1M to $2.11/1M $0.09495/1M to $10.55/1M confidential 3685 ms TTFT nebius · 99.60% available · n=1231
DeepSeek: DeepSeek V4 Flash 0423deepseek/deepseek-v4-flash
+14
27 routes
1,048,576 $0.07174/1M to $0.364144/1M $0.17724/1M to $0.842312/1M E2EE 5 cited scores 4807 ms TTFT deepseek · 114 tok/s · 99.59% available · n=986
MoonshotAI: Kimi K3moonshotai/kimi-k3
+20
39 routes
1,048,576 $2.05725/1M to $4.22/1M $10.28625/1M to $21.1/1M E2EE 10623 ms TTFT kimi · 55 tok/s · 100.00% available · n=427
GLM 5.2 Fast on Fireworksz-ai/glm-5.2-fast
4 routes
1,048,576 $2.2155/1M $6.963/1M ZDR 3411 ms TTFT fireworks · 149 tok/s · 100.00% available · n=218
Z.ai: GLM 5.3 Flashz-ai/glm-5.3-flash
+23
42 routes
1,048,576 $0.07385/1M to $2.11/1M $0.2532/1M to $10.55/1M E2EE 5540 ms TTFT zai · 97.50% available · n=126
Z.ai: GLM 5.3z-ai/glm-5.3
+23
42 routes
1,048,576 $1.0339/1M to $5.1273/1M $3.2494/1M to $12.81825/1M E2EE 3312 ms TTFT zero-g · 94 tok/s · 100.00% available · n=85
DeepInfraPolicy varies 101 models 2839 ms p50 · n=295
Fireworks AIPolicy varies 19 models 3411 ms p50 · n=265
GMI CloudPolicy varies 75 models 2841 ms p50 · n=28
Novita AIPolicy varies 98 models 2425 ms p50 · n=26
SiliconFlowPolicy varies 28 models 2268 ms p50 · n=496
VenicePolicy varies 58 models 2512 ms p50 · n=33
Nebius Token FactoryZDR 21 models 3685 ms p50 · n=500
BasetenZDR 17 models 2123 ms p50 · n=494

Browse every modelReview provider policiesOpen the full leaderboardSnapshot 2026-09-15T12:15:23.994Z

Questions

Does using a Chinese model send my prompts to China?

Only if you route it to a provider operated from China. The model's origin and the endpoint's operator are separate facts in the catalog. Most of these are open-weights models that US-registered providers also serve; a request routed to one of those reaches that provider's endpoint and is not sent on to the originating lab. A request to the vendor's own API does reach a China-registered operator, which is what the provider table on this page lists.

How do I use a Chinese model without China-bound traffic?

Set provider.jurisdiction to "us" on the request. The router then considers only providers whose recorded operator country is the United States, and fails closed when none serve that model, rather than silently routing elsewhere. provider.only pins an exact operator allowlist when a country test is not specific enough.

Which providers are operated from China?

The provider table on this page is generated from the catalog, listing every provider whose operator is a China-registered company according to that company's own privacy policy, terms, or regulatory filing. Some familiar Chinese AI brands are not on it: the api.z.ai and api.siliconflow.com services are operated by Singapore-registered companies under Singapore law per their own terms, so they are recorded under Singapore while the weights they serve stay grouped under the labs that built them.

Are Chinese models worse or less safe?

That is not a question a jurisdiction directory can answer, and this page does not try to. Benchmark scores, prices, and measured latency per route are on each model page and the live leaderboard. What the catalog records is the origin country and the operator country, each with the source it was read from, so a procurement review can weigh them alongside measurements rather than instead of them.

Workspace access

Sign in

Choose a sign in method to access your TrustedRouter workspace.

By signing in you agree to the terms of service and privacy policy.