Your cookie choices

We use necessary cookies to make this site work. With your permission, analytics cookies help us understand usage and marketing cookies support advertising.

You can reject non-essential cookies and still use the site. Change your choices anytime in Cookie settings.

Cookie settingsCookie policy

Chutes review

Updated September 28, 2026 · DeployCue Editorial Team · Facts checked September 26, 2026

We earn a commission when you sign up through links on this site. Partner status can affect where a brand appears in our rankings. Partners with a live offer in your country come first, then every other brand by score.

Our verdict

Chutes suits developers who want private, OpenAI-compatible inference at a low entry fee, with input from $0.025 to $3 and output from $0.098 to $15. Support runs on email, help centre and community, so teams needing a wide catalogue or phone help should look elsewhere.

Key facts

Best for
Private inference on GPUs you only pay for while a model runs.
Input price from
$0.0245 to $3↗
Output price from
$0.0978 to $15↗
Models offered
Unsloth/Mistral-Nemo-Instruct-2407-TEE, Nemotron-3-Nano-Omni-30B-TEE, Qwen/Qwen3-32B-TEE, Google/gemma-4-31B-turbo-TEE, Qwen/Qwen3-235B-A22B-Thinking-2507-TEE, Qwen/Qwen3.8-27B-TEE, Qwen/Qwen3.6-27B-TEE, Deepseek-AI/DeepSeek-V4-Flash-0731-TEE, Qwen/Qwen3.5-397B-A17B-TEE, Moonshotai/Kimi-K2.6-TEE, Deepseek-AI/DeepSeek-V3.2-TEE, Zai-org/GLM-5.1-TEE, Zai-org/GLM-5.2-TEE, Moonshotai/Kimi-K3-TEE↗
Largest context window
1,000,000↗
OpenAI-compatible API
Yes↗
Compliance certifications
SOC 2 Type II, GDPR, CCPA, TEE (Intel TDX, NVIDIA Protected PCIe)↗
Legal entity
Chutes Global Corp↗
Headquarters
Saint Kitts and Nevis↗
Support channels
Email, Help-centre, Community↗
Official website
chutes.ai

GPU cloud

GPU models
T4, V100, A6000, L40, A100, H100, H200↗
GPU hour from
$1.80↗
Billing granularity
Per second↗
Egress fee
Per GB of data transfer↗
Locations
6↗
Interconnect
VPC peering for enterprise customers↗
Pricing model
Hourly billing

Facts checked September 26, 2026, 5 sources

Apps and profiles

Elsewhere
Xgithub.com

Pricing

  • Chutes Qwen3 32B (USD)

    input-tokens: $0.10 / million

  • Chutes Qwen3 32B (USD)

    output-tokens: $0.42 / million

  • Chutes Qwen3 235B A22B (USD)

    input-tokens: $0.30 / million

  • Chutes Qwen3 235B A22B (USD)

    output-tokens: $1.20 / million

Pros and cons

Pros

  • Input fees run $0.025 to $3 and output $0.098 to $15, below the €0.07 and €0.23 across the group.
  • Billing is per second, and GPU hours start at $1.80 against a €0.63 group median.
  • Private deployment is covered by SOC 2 Type II, GDPR and CCPA, including TEE on Intel TDX and NVIDIA Protected PCIe.
  • The OpenAI-compatible API lets existing client code point at a different endpoint.

Cons

  • The catalogue is limited to the TEE models in unsloth/Mistral Nemo Instruct 2407 TEE, Nemotron 3 Nano Omni 30B TEE and Qwen/Qwen3 32B TEE, so most hosted frontier models are not served here.
  • GPU hours start at $1.80, the €1.75 level in the group, so cheap accelerators are thin on the list.
  • Support runs on email, help centre and community, with no phone line for production incidents.
Chutes

Ready to try Chutes?

Scored 9.3 / 10 by DeployCue.

See alternatives

Who it suits

  • Developers already running an OpenAI-compatible client who want inference to run in confidential computing on hardware they do not own.
  • Small teams that want to pay only while a model is answering, with per-second billing keeping short jobs cheap overall.
  • Organisations that must show where their data went, since TEE attestation and SOC 2 Type II appear in the certification list.

Who should look elsewhere

  • Teams wanting a broad hosted catalogue of named frontier models, who would find far more choice among Azure AI Foundry or Supascale AI listings.
  • Buyers shopping purely on the lowest entry input fee, since Salad starts at $0.06 and carries a different accelerator set.

Where it sits among rivals

On entry input price, Chutes ranks 17 of 71, and on entry output price 22 of 70. Its largest context window, 1,000,000, sits at 21 of 72 and matches the 1,000,000 group median. Prices cluster around the group middle, so the differentiator is private hardware, not cost.

Who runs it

The service is operated by Chutes Global Corp, registered in Saint Kitts and Nevis. Support runs on email, help centre and community, so day to day questions go to the help centre, by email or into the community, and there is no phone line. GPU work runs across 6 locations, with VPC peering for enterprise customers available for larger customers.

How Chutes compares

Frequently asked questions

What does it cost to run a model on Chutes?+

Entry input fees run $0.025 to $3 and entry output fees $0.098 to $15. Raw GPU work is billed per second, with hourly rates starting at $1.80. Plans are hourly, so there is no annual lock in.

Which models can I call?+

The catalogue in unsloth/Mistral Nemo Instruct 2407 TEE, Nemotron 3 Nano Omni 30B TEE and Qwen/Qwen3 32B TEE covers Qwen, DeepSeek, Kimi, GLM, Mistral and Gemma builds, each served in a TEE variant. It is a focused list rather than a full hosted catalogue, so check the model you need is present before you build on it.

How large a context can I use?+

The largest context window is 1,000,000, which matches the 1,000,000 group median and ranks 21 of 72. That is enough for long documents and large codebases in a single request.

Is the API a drop-in replacement?+

Yes, the OpenAI-compatible API is there, so an existing OpenAI client can point at a different base URL and send the same request format. Model names and pricing still differ, so responses and cost estimates should be checked per model.

How is private inference handled?+

Models run inside trusted execution environments, listed in SOC 2 Type II, GDPR and CCPA as TEE on Intel TDX and NVIDIA Protected PCIe, so weights stay inside the enclave while they are served. The same list records SOC 2 Type II, GDPR and CCPA coverage.

What does it charge for sending data out?+

Egress is charged Per GB of data transfer, on top of model and GPU fees. The group median for outbound transfer is €0, so returning large volumes of data can add up on a busy workload.

Where do the GPUs run?+

Capacity is spread across 6 locations, against a 4 group median, and the accelerator list in T4, V100 and A6000 runs from T4 and V100 up to A100, H100 and H200. Larger customers can ask about VPC peering for enterprise customers for direct VPC peering.

Who can get help?+

Support runs on email, help centre and community: the help centre, email and the community. There is no phone line, so teams that rely on calling a contact during an incident should plan an escalation path of their own.

Sources

  • chutes.ai/docs/getting-started/quickstart checked September 26, 2026

    Backs: OpenAI-compatible API

    “OpenAI-compatible API endpoints”

  • chutes.ai/docs/help/faq checked September 26, 2026

    Backs: Compliance certifications, Egress fee, GPU models, Interconnect, Locations

    “SOC 2 Type II compliance ... GDPR and CCPA compliant”

  • chutes.ai/help checked September 26, 2026

    Backs: Support channels

    “Use the fastest path for your question: docs for implementation details, Discord for quick community help, or email for account and billing support.”

  • chutes.ai/pricing checked September 26, 2026

    Backs: Billing granularity, Largest context window, GPU hour from, Input price from, Pricing model, Models offered, Output price from

    “deploy your own chute on a confidential GPU and pay by the second”

  • chutes.ai/terms checked September 26, 2026

    Backs: Headquarters, Legal entity

    “These Terms are governed by the laws of Nevis, Saint Kitts and Nevis”

Compiled from 5 source pages, checked September 26, 2026. We did not open an account.

Chutes

Chutes

Private inference on GPUs you only pay for while a model runs.

9.3/10See alternatives