Chutes Review 2026: Serverless AI Inference on Bittensor

Chutes is a decentralized, serverless inference platform for open-source AI models, run by Chutes Global Corp. It lets developers call models through an OpenAI-compatible API or deploy their own code, without managing GPUs. DefiLlama files it under Decentralized AI on the Bittensor chain, with the subnet token SN64.

The homepage lists LLMs from families such as Kimi, GLM, DeepSeek, Qwen, Gemma, MiniMax and Mistral, plus image, video, speech, music, embedding and moderation models. It also lists partners and integrations such as OpenRouter, Kilo, FetchAI and Phala, and runs consumer apps including Chutes Chat and fictio.

DefiLlama shows about $46.9 million in TVL, down 7.3% over seven days, and no audits listed. The project has been listed since February 2026. The platform's claims about scale, such as trillions of tokens per month, come from its own site.

Chutes is a serverless layer for running AI models. Developers either call a hosted model through the API or package their own code as a 'chute' and let the platform run it. They do not rent or manage GPU servers themselves, and capacity is described as always on, without cold starts for popular models.

The compute is decentralized and runs on Bittensor, where it is tracked as the SN64 subnet. In a Bittensor subnet, independent operators supply resources and are rewarded through the network's token system. The project does not detail how Chutes assigns work or rewards providers, so those mechanics should be checked in its documentation.

For privacy, some models run in a trusted execution environment, a hardware-isolated area meant to keep prompts and outputs hidden from the machine's operator. Users pay from a topped-up balance based on tokens processed or GPU seconds.

Chutes uses pay-as-you-go pricing. You top up a balance and pay only for tokens and GPU time used, with no subscription or minimum commitment. Private chutes are billed per second of GPU time.

Per-token rates are published per model on the site. Examples shown on the homepage per million tokens include DeepSeek V4.1 Flash at $0.30 input and $1.20 output, and GLM 5.2 Fast at $2.45 input and $7.70 output, with cached input priced lower. Rates change by model, so use the live pricing page and cost estimator before committing. Enterprise pricing is custom and not published.

DefiLlama lists no audits for Chutes, and public sources describe no formal security assessment. The site does link a vulnerability reporting channel and legal documents, including privacy terms and a data processing agreement. No hacks or incidents appear.

The main risks are counterparty and privacy risks. Prompts are processed on decentralized infrastructure, so confidentiality depends on the TEE option and the platform's own controls. Some forum users questioned the platform's security and privacy, so avoid sending sensitive data to non-TEE models without checking the terms. Keep prepaid balances modest, since this is a young platform. The SN64 token also carries market risk through Bittensor.

Chutes is for developers, startups and AI hobbyists who want to run open-source models through an API without managing GPUs, and for people who follow Bittensor subnets. It is less suitable for teams that need audited, regulated infrastructure or guaranteed data residency without extra due diligence.