Skip to content
VoltageGPU
Contact
Deploy
EU · GDPR Art. 28 · Intel TDX · OpenAI-compatible

VoltageGPU vs Together AI

Together AI is operated by Together Computer, Inc., a US Delaware corporation headquartered in San Francisco. VoltageGPU (VOLTAGE EI, Solaize, France, SIREN 943 808 824) is not affiliated with Together Computer, Inc.

Same open-weight models, same OpenAI-compatible API, attestation is the only thing they do not ship. Together AI is fast, broad, and cheap on US multi-tenant inference. VoltageGPU runs a smaller catalogue of open-weight models inside Intel TDX enclaves under a French controller, with an attestation of the serving instances; our API relay sees prompts in clear. Different products, not just different prices.

Requests pass through our relay, which sees them in clear (TLS ends at our host). To keep plaintext off our servers, run your own inference server in a Confidential VM and terminate TLS inside its trust domain. Confidential VMs

Headline pricing

Per-million-token list price by model tier. VoltageGPU rows are TEE models (Intel TDX), priced from our live catalog. A dash means the competitor does not publish a comparable SKU. Live prices on /pricing.

TierVoltageGPU (TEE)Together AI
Mid-size open (30B class)
Qwen3.8-27B-TEE
in $0.0700 · out $2.31 / 1M tok
Qwen 2.5 72B Instruct
in $1.20 · out $1.20 / 1M tok · no TEE, US multi-tenant
Fast mid-size (70B turbo)
gemma-4-31B-turbo-TEE
in $0.2200 · out $0.6800 / 1M tok
Llama 3.3 70B Instruct Turbo
in $0.8800 · out $0.8800 / 1M tok · no TEE, 200–400 tok/s, US multi-tenant
Frontier MoE
Qwen3.5-397B-A17B-TEE
in $0.8300 · out $5.55 / 1M tok
DeepSeek V3
in $1.25 · out $1.25 / 1M tok · no TEE, cheaper on output, US multi-tenant
Confidential techIntel TDX + trust-domain GPU isolationNot offered (no Intel TDX, no GPU TEE on inference)
AttestationIntel DCAPNone
BillingPer-token, OpenAI-compatiblePer-token, OpenAI-compatible; dedicated endpoints hourly
OperatorVOLTAGE EI (France)Together Computer, Inc. (US, Delaware), San Francisco HQ
Setup~30 sec, base URL and API key~30 seconds (API key issuance)
JurisdictionFrench company, GDPR Art. 28; compute at US sub-processors (SCCs)US company

Same models, same API, attestation is the structural delta

Together AI and VoltageGPU sit closer to each other on the product map than most of our other comparison pages. Both are OpenAI-compatible inference platforms, change the base_url from api.openai.com/v1 to api.together.xyz/v1 or api.voltagegpu.com/v1 and existing client code keeps working. Both serve open-weight models from overlapping families (Qwen, DeepSeek, Gemma). Both bill per token with no minimum commit. From a developer-integration point of view the two platforms are interchangeable in a thirty-second swap of an environment variable. Requests pass through our relay, which sees them in clear (TLS ends at our host). To keep plaintext off our servers, run your own inference server in a Confidential VM and terminate TLS inside its trust domain.

The product-shape delta is what happens to a prompt after it crosses the API boundary. On Together AI the prompt enters a US multi-tenant inference stack operated by Together Computer, Inc., a Delaware corporation headquartered in San Francisco, running on standard hypervisors with SOC 2 Type II controls. There is no Intel TDX VM around the inference worker, no GPU TEE on the H100/H200 silicon, no Intel DCAP attestation quote that an auditor can re-verify offline, and no European operator on the contract. That is sufficient for the majority of inference workloads and Together has built one of the strongest brands in the open-weight inference category on exactly that posture.

On VoltageGPU the same prompt reaches our relay, which forwards it in clear to a model running inside an Intel TDX enclave at our inference provider, whose memory the host cannot read. GET /v1/attestation returns the TDX quotes and NVIDIA GPU evidence of the serving instances against the Intel root certificate; it proves those instances are TDX guests, not that a given response came from them, and it is not issued with each call. The operator is VOLTAGE EI in Solaize, France, registered under SIREN 943 808 824, and the contract is signed against the French legal framework.

SOC 2 on standard silicon vs a model in a hardware enclave

Together AI publishes a SOC 2 Type II attestation and operates a Data Processing Addendum for enterprise customers. That is the standard compliance posture every serious US inference platform exposes and it is sufficient for a large fraction of enterprise workloads, internal copilots, public-content summarisation, code generation against non-sensitive repositories, marketing-content pipelines, evaluation suites, and the long tail of AI work where the inputs were already non-sensitive when they reached the API. For those workloads Together is a price-competitive, broad-catalogue, fast-Turbo-tier inference platform and there is no good reason to pay more.

The posture stops being sufficient for buyers who want hardware evidence about who can read prompt memory. A SOC 2 audit confirms organisational controls; it does not constrain the host operator at the silicon layer. If a US administrator with legitimate console access wants to introspect a running inference worker, for legitimate operational reasons or under US legal compulsion through the Cloud Act framework, the standard hypervisor permits it. For prompts containing client files protected by French bar-association secrecy, health data under HDS scope, cardholder data under PCI DSS, or any personal data covered by GDPR Article 9, some buyers want the technical measure enforced by hardware, not only promised by contract.

VoltageGPU addresses part of that, and not by geography. The data does leave the EU, to compute capacity operated by documented sub-processors (/legal/subprocessors). Our relay sees each prompt in clear while forwarding it; the model then runs inside an Intel TDX enclave whose memory the host cannot read, and the TDX quote of the serving instances is signed by Intel and verifiable offline. The operator is a French entity and the GDPR Article 28 DPA is signed against the French legal framework. Together AI does not offer a TDX enclave for the model today and has not announced TDX inference; that is not a Together bug, it is the inference category they chose to build.

Where Together wins, and it wins on real ground

It would be dishonest to write a comparison page against Together AI without acknowledging the categories where they are clearly the better tool. Catalogue breadth is the most obvious: Together serves 200+ open-source models including image generation (FLUX.1 dev, Stable Diffusion XL), video generation (Wan, Hunyuan), audio, embeddings, rerankers, and safety classifiers. VoltageGPU lists a smaller set of TEE text and code models (live list on /models), Qwen, Gemma 4, DeepSeek, GLM and Kimi, and does not currently ship a confidential image-or-video inference product. For any workload that needs FLUX, Wan, or Stable Diffusion at API speed, Together is the right answer and VoltageGPU is not in that category at all today.

On raw per-token economics Together wins outright on the tiny-model and frontier-output tiers. Llama 3.1 8B Instruct Turbo is $0.18 / $0.18 per million tokens on Together, we do not list an 8B TEE SKU, so for high-volume tiny-model inference where the workload is non-confidential, Together is the price-correct choice. More importantly, DeepSeek V3 is $1.25 / $1.25 flat on Together; compare it with our Qwen3.5-397B-A17B-TEE frontier tier in the live table above, where output costs more than input. For an output-heavy frontier workload that does not need TEE, long-form generation, code synthesis at scale, agent-loop reasoning where output tokens dominate, Together wins the economic comparison and the buyer should choose Together.

Together also ships a dedicated-endpoint product starting at $3.99/hr for an H100 80GB that lets enterprise customers buy predictable throughput rather than burst per-token capacity. That is a category VoltageGPU has not built today on the inference side, our confidential GPU pods exist, but the "dedicated managed inference endpoint with auto-scaling" shape is a Together product feature we do not match. Turbo-tier latency on Llama 3.3 70B at 200–400 tokens per second is industry-leading and not a number we publish on our confidential serving infrastructure.

The honest summary is product-shape, not provider-quality: Together is the breadth-and-speed king on US multi-tenant open-weight inference and remains the right answer for unconstrained workloads on the broadest model catalogue. VoltageGPU is confidential inference from a French operator and is the right answer when the client contract or the threat model calls for the model to run in a hardware enclave. A workload that picks the wrong-shape tool will be unhappy with both.

FAQ

Is Together AI GDPR compliant?

Together AI is SOC 2 Type II certified and operates a Data Processing Addendum for enterprise customers, which together satisfy the formal Article 28 requirement of a written processor agreement. For the majority of inference workloads, internal copilots, public-content summarisation, code generation, non-sensitive enterprise pipelines, that posture is sufficient and Together is a credible choice. It is not always enough where the workload involves personal data under GDPR Article 9 (health, biometrics, sex life, trade-union membership), client files protected by professional secrecy, or processing that triggers the EU AI Act high-risk classification, where buyers often want the technical measures backed by hardware isolation. Together's inference runs on standard multi-tenant infrastructure operated from the US, the data residency option is limited and the silicon does not produce attestation evidence. VoltageGPU adds a model in Intel TDX under a French controller and a DPA; our relay still sees prompts in clear, and we hold no SOC 2 (see /trust).

Does Together AI offer EU data residency or EU regions?

Together AI's inference platform is operated primarily from US infrastructure as of May 2026 and the company has not publicly shipped a generally-available EU inference region. Enterprise customers can negotiate data-handling terms through the DPA, but EU residency on Together inference is not a self-service configuration the way it is on AWS Bedrock or Azure OpenAI EU regions. VoltageGPU does not offer EU residency for inference either: our operator entity is French (VOLTAGE EI, SIREN 943 808 824), but the inference provider is in the US (listed on /legal/subprocessors); what we add is the model inside Intel TDX, with an attestation of its instances. For a European buyer whose data is non-sensitive and where SOC 2 + DPA is contractually acceptable, Together is a viable option even from US infrastructure and frequently the price-correct choice.

Which is cheaper, VoltageGPU or Together AI?

It depends on the model tier and the input/output ratio. Together lists Qwen 2.5 72B at $1.20 / $1.20, Llama 3.3 70B Turbo at $0.88 / $0.88, DeepSeek V3 at $1.25 / $1.25 and Llama 3.1 8B Turbo at $0.18 / $0.18 per million tokens; our TEE models are priced in the live table above. Input-heavy workloads tend to come out cheaper with us, output-heavy ones on Together, and we do not list a TEE 8B model. The right framing is not "which is cheaper" it is "which tier matches the workload and does it need attestation".

Can I use Together AI for HIPAA or regulated workloads?

Together AI has built enterprise compliance infrastructure including SOC 2 Type II and discusses BAA availability with healthcare customers on request, which covers the formal contractual baseline US covered entities expect. That covers the legal/regulatory framework on the contract side. The technical pattern still relies on standard multi-tenant inference infrastructure: the contract promises the workload data will not be accessed inappropriately, the silicon does not enforce it. For PHI, the architectural alternative is a model in a hardware enclave: VoltageGPU runs it in Intel TDX at our US inference provider, our relay sees requests in clear, and we offer no BAA yet and hold no HDS certification (see /trust). Together and VoltageGPU therefore sit at adjacent tiers: a US healthcare buyer with a contractual BAA framework on non-critical PHI may find Together sufficient; a buyer who wants the model's host constrained by hardware can look at VoltageGPU, within those limits.

How hard is it to migrate from Together AI to VoltageGPU?

For OpenAI-compatible inference workloads the migration is a same-day exercise. Both platforms expose the same OpenAI SDK surface for chat completions, function calling and streaming (embeddings answer 503 at VoltageGPU, where no TEE embedding model is served), and the only client-code change is the base_url and the API key. Move from base_url="https://api.together.xyz/v1" to base_url="https://api.voltagegpu.com/v1", swap the bearer token, and existing Python or TypeScript SDK code keeps working. Requests pass through our relay, which sees them in clear (TLS ends at our host). To keep plaintext off our servers, run your own inference server in a Confidential VM and terminate TLS inside its trust domain. The structural caveats are model-name remapping (Together's "meta-llama/Llama-3.3-70B-Instruct-Turbo" maps to a different model identifier on our catalogue, and we expose TEE-suffixed identifiers such as google/gemma-4-31B-turbo-TEE rather than the bare HuggingFace path) and product-feature gaps (we do not currently match Together's image-generation models, dedicated managed endpoints, or 8B Turbo tier). For text-and-code inference on the mid-size-and-up tiers, the migration is typically that endpoint switch plus a model-name remap. For image/video generation or dedicated-endpoint workloads, VoltageGPU is not the right destination today and the workload should stay on Together.

Other inference APIs compared

Breadth and speed or hardware attestation, pick the right tool

Together AI is the right answer for unconstrained workloads on the broadest open-weight catalogue. VoltageGPU is the right answer when you want the model in a hardware enclave. Drop-in OpenAI-compatible: change the base_url. Requests pass through our relay, which sees them in clear (TLS ends at our host). To keep plaintext off our servers, run your own inference server in a Confidential VM and terminate TLS inside its trust domain.

Get an API keyArchitecture & attestation