Back to research
Data as of 10 min read

Riding token subsidies past 3T monthly tokens, then halving free volume to lift GPU revenue 44.9%

Chutes.ai is a decentralized, OpenAI-compatible serverless inference API originated by Bittensor subnet team Rayon Labs, pairing cheap open-weight LLMs with TEE-protected confidential compute across $10/$20 monthly tiers and PAYG billing. Launched alongside DeepSeek R1 in Jan 2025 via generous free daily quotas and OpenRouter routing, it surged past 3T monthly tokens by Jun 2025 and 34T cumulative tokens by Sep 2026—before intentionally cutting token volume in half in Mar 2026 to boost revenue per token (+37.7%) and per GPU (+44.9%).

Who it is for
Token-hungry AI roleplay builders (SillyTavern/JanitorAI users), budget-conscious coding agent developers (OpenCode, Kilo, Cursor power users), privacy-minded TEE compute buyers, and Bittensor-native devs.
How it charges
Dual-track subscription and usage billing: Plus at $10/mo (2,000 req/day), Pro at $20/mo (5,000 req/day), plus pay-as-you-go rates (e.g., DeepSeek V4 Flash at $0.44/$1.32 per 1M tokens; evolved from early free tiers and an Aug 2025 one-time $5/200-req-day transition).
01

Executive Summary

What this business is

Customers

Token-hungry AI roleplay builders (SillyTavern/JanitorAI users), budget-conscious coding agent developers (OpenCode, Kilo, Cursor power users), privacy-minded TEE compute buyers, and Bittensor-native devs.

Offering

Decentralized serverless inference and compute API (OpenAI-compatible) for open-source AI models: lets developers deploy custom workloads ('chutes'), highlights TEE confidential compute, and runs distributed training (e.g., training an 8B model for $6,500 across 240 consumer GPUs in 13 countries in Sep 2026).

Business model

Subnet-subsidized compute flywheel evolving into margin-disciplined API monetization: leveraged Bittensor emissions to win massive initial token volume on OpenRouter, then pruned unprofitable free load in favor of paid subscriptions, PAYG usage, and TEE privacy premiums.

Pricing

Dual-track subscription and usage billing: Plus at $10/mo (2,000 req/day), Pro at $20/mo (5,000 req/day), plus pay-as-you-go rates (e.g., DeepSeek V4 Flash at $0.44/$1.32 per 1M tokens; evolved from early free tiers and an Aug 2025 one-time $5/200-req-day transition).

Takeoff mode
Upstream Aggregator & Token Subsidy Flywheel
Pricing Structure
Plus $10/mo (2K req/d) · Pro $20/mo (5K req/d) + PAYG
Company & Roots
Chutes Global Corp (~8 staff) · ex-Rayon Labs / Bittensor

Launching in late January 2025 just days after DeepSeek R1 stunned the developer world, Chutes filled an immediate availability and pricing vacuum. Backed by decentralized Bittensor miners, it stood up free DeepSeek V3/R1 endpoints directly and inside OpenRouter (~200 free daily calls). That instant-gratification wedge ignited the two heaviest open-source token consumers on the web—AI roleplayers in SillyTavern and budget coding-agent tinkerers—exploding throughput 250x by June 2025.

Chutes' pricing evolution offers a textbook lesson in curing AI subsidy hangovers. Free daily tiers drove 3T+ tokens/month by June 2025, yet agentic coding sessions burning $30 in two hours at ~$0.20/1M tokens forced discipline: introducing a $5 threshold in Aug 2025, transitioning into $10/$20 daily-capped plans plus transparent PAYG, and intentionally cutting token volume in half in March 2026. Pruning subsidized freeloaders lifted revenue per token +37.7% and revenue per GPU +44.9%.

Operated by Chutes Global Corp (~8-person lean team) spun out of Bittensor subnet builder Rayon Labs (Sep 2024). Core backend engineering is led by Jon Durbin (creator of Airoboros), supported by dedicated Head of GTM and Growth Burhan Khan and Reddit community operator Mikesoft (@thestreamcode). Internal traffic shows a US-dominant user base (55%).

Preparing your reading

A work in progress.

2026.11.08 · V0.4

Refined product categories and extension rankings. Added one-line Agent setup and three membership options.