Qwen/Qwen3-32B-TEE
32B parameter model optimized for reasoning, coding, and instruction following. 40K context window. Most popular TEE model with 33.2M runs in 7 days.
Pricing: $0.15 per million input tokens, $0.44 per million output tokens
Confidential AI Models, TEE Inference
14 TEE models (Trusted Execution Environment) running inside Intel TDX hardware enclaves. Prompts and responses are encrypted in memory using AES-256 with CPU-fused keys, even the infrastructure operator cannot read them. OpenAI-compatible REST API. Drop-in replacement for OpenAI and Anthropic for regulated workloads.
What are TEE models?
TEE models run inside an Intel TDX Trusted Domain, a hardware-isolated virtual machine verified by CPU microcode. This is the same confidential computing technology used by Microsoft Azure Confidential VMs and Google Cloud Confidential VMs. Every prompt and response is encrypted in memory during processing, making TEE models the correct choice for legal documents, patient records, financial data, and any workload under GDPR Article 28, HIPAA, DORA, or professional secrecy rules.
Popular TEE Models
Qwen/Qwen3.5-397B-A17B-TEE
397B parameter mixture-of-experts model. 256K context window for long document processing, entire contracts or financial reports in one pass.
deepseek-ai/DeepSeek-V3.2-TEE
Latest DeepSeek model with strong performance on code and reasoning tasks.
Pricing: $0.20 per million input tokens, $0.89 per million output tokens
Qwen/Qwen3-32B-TEE
Reasoning model with chain-of-thought capabilities. Suitable for CFA-grade financial analysis, multi-step legal reasoning, and complex compliance analysis.
Pricing: $0.46 per million input tokens, $1.85 per million output tokens
moonshotai/Kimi-K2.6-TEE
4.6M runs in 7 days. Strong instruction following.
moonshotai/Kimi-K2.6-TEE
Moonshot AI model with large context window support.
mistral-small-24b-tee
24B parameter Mistral model. Efficient and capable, ideal for cost-sensitive workloads.
Pricing: $0.06 per million input tokens, $0.18 per million output tokens
zai-org/GLM-4.6-TEE, GLM-5-TEE, GLM-5.1-TEE
ZhipuAI GLM models with bilingual (Chinese/English) support.
openai/gpt-oss-120b-TEE, gpt-oss-20b-TEE
OpenAI open-weight models running inside Intel TDX enclaves.
Other TEE Models
Also available: DeepSeek-V3.1-TEE, DeepSeek-V3.1-Terminus-TEE, DeepSeek-V3-0324-TEE, DeepSeek-TNG-R1T2-Chimera-TEE, GLM-4.7-TEE, deepseek-ai/DeepSeek-V4-Flash-0731-TEE, Qwen3-Coder-Next-TEE.
API Compatibility
All TEE models are accessible via an OpenAI-compatible REST API. Change your base URL to https://api.voltagegpu.com/v1 and use your VoltageGPU API key. Compatible with OpenAI SDKs (Python, Node.js, Go), LangChain, LlamaIndex, CrewAI, OpenClaw, and any other OpenAI-compatible client library.
curl https://api.voltagegpu.com/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "Qwen/Qwen3-32B-TEE",
"messages": [{"role": "user", "content": "Analyze this contract..."}]
}'Security Architecture
- Intel TDX, hardware-isolated Trust Domains verified by CPU microcode
- AES-256 memory encryption with CPU-fused keys (hypervisor cannot decrypt)
- trust-domain GPU isolation, encrypted CPU-to-GPU channel, no bus snooping
- Hardware attestation, cryptographic proof of enclave integrity
- Per-token model verification, cryptographically proves output came from declared model
- Zero data retention, prompts and responses purged when session ends
Compliance
TEE models satisfy technical safeguards for GDPR Article 28, HIPAA, SOC 2 (audit in progress), DORA, NIS2 Directive, and French CNIL guidelines. Law firms, accounting firms, clinics, and fintech companies use VoltageGPU TEE models to analyze sensitive documents without violating professional secrecy or regulatory constraints.