OpenAI compatible API · Attested · Public status

GLM-5 & GLM-5.2 API — Private, Attested Inference

OpenAI-compatible API for GLM-5 and GLM-5.2 on attested hardware. ZDR and TEE tiers, no z.ai account needed, and privacy you can verify with one curl.

Verify gateway

Onebase URL to migrate

100sof models and routes

0prompt or output logs. Always.

GLM-5 & GLM-5.2 API

Run GLM-5 and GLM-5.2 on attested hardware without sending your prompts to the model vendor.

TrustedRouter serves glm-5, glm-5.2, and glm-5.2-fast through an OpenAI-compatible API. Change one line, the base URL, and keep your SDK, your model ids, and your code. You do not need a z.ai account.

Every GLM route here is served by non-Chinese hosting providers on attested infrastructure. Zero-Data-Retention and TEE tiers are available, and on TEE routes even the hosting provider cannot read your prompt. TrustedRouter never logs prompt or output content. Ordinary synchronous and streaming inference does not retain it. The opt-in Batch API temporarily retains enclave-encrypted artifacts for up to 30 days. Pay with prepaid credits, bring your own key, or pay per request with x402. Continue with MetaMask works without a traditional account.

Open the playground See GLM pricing

glm-5.2 via the OpenAI SDKPython

from openai import OpenAI

client = OpenAI(
    base_url="https://api.trustedrouter.com/v1",
    api_key=TR_KEY,
)
r = client.chat.completions.create(
    model="z-ai/glm-5.2",
    messages=[{"role": "user", "content": "Refactor this function."}],
)
print(r.choices[0].message.content)

Model routes

Three GLM routes, one API

glm-5, glm-5.2, and glm-5.2-fast are all live, alongside 220+ model routes across 30+ providers. When a hosting provider fails, requests roll over automatically. Kimi, DeepSeek, and Qwen run the same way; see Chinese open models served without the vendor and measured latency data from real routed requests.

Privacy tiers

ZDR and TEE on the prompt path

Pick Zero-Data-Retention routes, where providers contractually keep nothing, or TEE routes, where end-to-end confidential compute means even the provider cannot read the prompt. Details at /security.

Measured behavior

Host-applied filtering, measured

We ran 60 sensitive questions against the same GLM weights twice. The vendor-hosted API answered 30. The attested TrustedRouter route answered all 60. The filtering is applied by the serving host.

z.aiTrustedRouter

Model accessDirect, first-party endpoint from the team that builds GLMglm-5, glm-5.2, and glm-5.2-fast among 220+ routes

Privacy verificationA privacy policy you take on trustLive hardware attestation, open-source gateway, published image digest

Prompt pathPrompts are processed on the model vendor's serversPrompts do not go to the model vendor; ZDR and TEE tiers

Sensitive questions, our 60-question test30 of 60 answered60 of 60 answered, same weights

Account and paymentVendor account sign-upPrepaid credits, BYOK, or x402 crypto with MetaMask sign-in

OutagesSingle providerAutomatic rollover across providers

Verify, don't trust

Same weights, different host, different answers.

The test was simple. Sixty sensitive questions, the same GLM weights, two routes. Through the vendor-hosted API, 30 of 60 came back answered. Through TrustedRouter's enclave, 60 of 60. The gap comes from filtering applied by the serving host, so choosing where the weights run decides what you get back.

You do not have to take the enclave on faith. The prompt path runs inside a hardware-attested confidential-compute gateway. The source is open, the image digest is published, and a curl to the attestation endpoint returns a JWT signed by the CPU vendor's root key, bound to your live TLS session so it cannot be replayed. Attestation beats a privacy policy. It also holds up under load: a document-processing company scaled from zero to over a billion tokens per week through the attested gateway within three weeks of onboarding. Check the attestation yourself at /security, or start with GLM-5.2 in the playground.

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.

554public models

95providers

1761configured routes

314ZDR routes

35provider E2EE routes

5539recent availability samples

Model	Providers	Context	Input	Output	Privacy	Measured route
GLM 5.2 Fast on Fireworks`z-ai/glm-5.2-fast`	4 routes	1,048,576	$2.2155/1M	$6.963/1M	ZDR	3050 ms TTFT fireworks · 153 tok/s · 100.00% available · n=171
Z.ai: GLM 5.2`z-ai/glm-5.2`	+24 44 routes	1,048,576	$0.7174/1M to $2.434307/1M	$1.5825/1M to $7.029062/1M	E2EE 4 cited scores	1806 ms TTFT baseten · 128 tok/s · 100.00% available · n=145
Z.ai: GLM 5.3 Flash`z-ai/glm-5.3-flash`	+23 41 routes	1,048,576	$0.07385/1M to $2.11/1M	$0.2532/1M to $10.55/1M	E2EE	8850 ms TTFT zai · 100.00% available · n=128
Z.ai: GLM 5.3`z-ai/glm-5.3`	+24 43 routes	1,048,576	$1.0339/1M to $5.1273/1M	$3.2494/1M to $12.81825/1M	E2EE	23042 ms TTFT fireworks · 94.12% available · n=88
Z.ai: GLM 5.1`z-ai/glm-5.1`	+8 20 routes	202,752	$1.0339/1M to $1.6247/1M	$3.2494/1M to $5.1062/1M	E2EE 5 cited scores	1122 ms TTFT friendli · 100.00% available · n=21
Z.ai: GLM 5`z-ai/glm-5`	+2 10 routes	202,752	$0.9073/1M to $1.055/1M	$2.69025/1M to $3.376/1M	ZDR 5 cited scores	3309 ms TTFT zai · 100.00% available · n=13

Fireworks AIPolicy varies 19 models 3050 ms p50 · n=371

BasetenZDR 17 models 1221 ms p50 · n=490

Z.AIPolicy varies 14 models 8850 ms p50 · n=78

DeepInfraPolicy varies 101 models 3830 ms p50 · n=494

FriendliAIPolicy varies 7 models 1331 ms p50 · n=31

GMI CloudPolicy varies 76 models 3851 ms p50 · n=16

Novita AIPolicy varies 98 models 2920 ms p50 · n=40

ParasailZDR 26 models 2781 ms p50 · n=29

Browse every model Review provider policies Open the full leaderboardSnapshot 2026-09-17T13:23:21.989Z

Questions

Which GLM models can I call on TrustedRouter?

TrustedRouter serves glm-5, glm-5.2, and glm-5.2-fast, alongside 220+ model routes across 30+ providers. The API is OpenAI-compatible: change base_url, keep your SDK, model ids, and code. If a hosting provider goes down, requests roll over to another automatically. You can try it in the playground at /chat before wiring up a key.

Do my prompts go to Zhipu or z.ai?

No. GLM routes on TrustedRouter are served by non-Chinese hosting providers on attested infrastructure, so prompts do not go to the model vendor. Zero-Data-Retention and TEE tiers are available, and on TEE routes even the hosting provider cannot read the prompt. TrustedRouter never logs prompt or output content. Ordinary synchronous and streaming inference does not retain it. The opt-in Batch API temporarily retains enclave-encrypted artifacts for up to 30 days.

Why does GLM answer more questions through TrustedRouter than on the vendor API?

In our 60-question test, the same GLM weights answered 60 of 60 sensitive questions through TrustedRouter's attested enclave versus 30 of 60 through the vendor-hosted API. That points to host-applied filtering: the restriction lives in the serving stack, not in the open weights. TrustedRouter's GLM routes are served on attested infrastructure without that serving-layer filter, so you see the model's own behavior.

How does GLM-5.2 API pricing work?

Per-model rates for glm-5, glm-5.2, and glm-5.2-fast are published at /pricing as a thin markup over provider list prices. You can pay with prepaid credits through Stripe, bring your own provider key, or pay per request with x402 crypto payments. Continue with MetaMask lets you sign in with a wallet, with no traditional account required.