Should I rely on -TEE in the model name?

No. Filter on the confidential_compute boolean from the live model object.

Does shape-valid evidence prove cryptographic attestation?

No. It shows the evidence object parsed to the expected shape. A cryptographic claim requires the DCAP validation path to run and pass.

Private inference

Make privacy claims from live evidence.

Chutes exposes confidential-compute metadata in the live model catalog. Use the boolean field, attach the verification date, and do not upgrade a parsed evidence response into a cryptographic attestation claim unless DCAP validation actually ran.

Browse live models Open TEE docs

Private inference draft TEE skill source System prompt TEE rules

Quick config

The minimum to point this client at Chutes. Every value below is rendered from the live catalog or is a stable endpoint fact — copy it and go.

TEE-only catalog filter

bash

curl https://llm.chutes.ai/v1/models

# In code, filter each item on:
# model.confidential_compute === true

Live catalog

13 listed models report confidential_compute=true, verified Jun 25, 2026.

Selection rule

Filter confidential_compute === true

Evidence nonce

64 hex characters, 32 bytes

Honesty boundary

Shape-valid is not a DCAP verdict

60-second setup

Read /v1/models

Use the catalog object that crawlers and users can verify, not a suffix in the model name.

Filter on the boolean

A TEE statement should be tied to confidential_compute=true and a verification date.

Fetch evidence carefully

Evidence endpoints need Bearer auth and a fresh 64-hex nonce.

Evidence is a separate step

Model metadata is public. Evidence endpoints are authenticated and chute-specific. Keep those two facts separate in UI copy and agent prompts.

Open evidence script

Evidence request shape

bash

export CHUTES_API_KEY=cpk_...
export CHUTE_ID=<chute_id>
export NONCE=<64-hex-character-nonce>

curl "https://api.chutes.ai/chutes/$CHUTE_ID/evidence?nonce=$NONCE" \
  -H "Authorization: Bearer $CHUTES_API_KEY"

What pages can say

Allowed

13 listed models report confidential_compute=true, verified Jun 25, 2026.

Requires live data

Count, prices, context, modalities, and TEE flags.

Allowed

Evidence parsed to the expected shape.

Not allowed without DCAP

Cryptographic attestation passed.

Live model facts

This table is server-rendered from the live API with a vendored snapshot fallback.

Live model menu

13 models. All report confidential_compute=true, verified Jun 25, 2026.

Live API

Model	Context	$/1M in	$/1M out	Features	Good at
unsloth/Mistral-Nemo-Instruct-2407-TEE	128K	$0.0245	$0.0978	No advertised feature flags	general chat
Qwen/Qwen3-32B-TEE	40K	$0.104	$0.416	JSON modeTool callingStructured outputReasoning	cost-aware tool loops
google/gemma-4-31B-turbo-TEE	128K	$0.12	$0.37	JSON modeTool callingStructured outputReasoning	vision-capable agents
MiniMaxAI/MiniMax-M2.5-TEE	192K	$0.15	$1.20	JSON modeTool callingStructured outputReasoning	cost-aware tool loops
Qwen/Qwen3-235B-A22B-Thinking-2507-TEE	256K	$0.2989	$1.1957	JSON modeStructured outputTool callingReasoning	long-context work
Qwen/Qwen3.6-27B-TEE	256K	$0.30	$2.00	JSON modeTool callingStructured outputReasoning	vision-capable agents
moonshotai/Kimi-K2.5-TEE	256K	$0.44	$2.00	JSON modeStructured outputTool callingReasoning	vision and video agents
Qwen/Qwen3.5-397B-A17B-TEE	256K	$0.45	$3.00	JSON modeTool callingStructured outputReasoning	vision-capable agents
moonshotai/Kimi-K2.6-TEE	256K	$0.66	$3.50	JSON modeStructured outputTool callingReasoning	vision and video agents
zai-org/GLM-5-TEE	198K	$0.95	$2.55	JSON modeStructured outputTool callingReasoning	long-context work
zai-org/GLM-5.1-TEE	198K	$0.98	$3.08	JSON modeStructured outputTool callingReasoning	long-context work
deepseek-ai/DeepSeek-V3.2-TEE	128K	$1.00	$1.00	JSON modeTool callingReasoningStructured output	reasoning and chat
zai-org/GLM-5.2-TEE	1.0M	$1.40	$4.40	JSON modeStructured outputTool callingReasoning	long-context work

Troubleshooting

Symptom

A model ID includes -TEE

Likely cause

Names are not the source of truth.

Fix

Use confidential_compute from the live model object.

Symptom

Evidence request returns 422 or 400

Likely cause

The nonce is missing or not exactly 64 hex characters.

Fix

Generate a 32-byte nonce and hex-encode it.

Symptom

A page claims cryptographic attestation

Likely cause

Only shape validation ran.

Fix

Say evidence parsed successfully unless DCAP verification passed.