Chutes is a model-access platform built on Bittensor (subnet 64) that sells per-token inference on open-weight models, optional monthly plans, and private GPU deployments. Its per-token prices were the lowest I verified in this category, and price is the main reason to pick it.
What it is #
Chutes serves open-weight models (Qwen, DeepSeek, Kimi, GLM, Mistral, Gemma) from hosted instances and charges by the million tokens, with no subscription required. Featured models run on confidential TEE compute, and billing accepts USD or TAO from a Bittensor wallet. A CLI can also deploy a private chute on self-serve GPU capacity billed per second, with a one-time deployment fee scaled to the GPU choice.
Status #
Active and shipping changes as of September 2026, with a live pricing page, status page, and regular community announcements. 2026 was a repair year for the business model: on February 27 the team retired the free 200-requests-per-day Early Access perk, capped subscriptions at 5x pay-as-you-go value, and cut frontier models from the $3 Base tier after its tables showed top users extracting up to 324x their subscription price. A March 20 update reports tokens served down 45% over six weeks while revenue per million tokens rose 37.7% since February 1, after ending roughly 10B tokens/day of free-quota serving to about 11,000 users. I could not verify funding or traffic figures.
Strengths #
- Lowest per-token list prices I found this run, for example DeepSeek-V3.2 at $1.00/$1.00 per 1M in/out and Mistral-Nemo at $0.0245/$0.0978.
- Many model families behind one OpenAI-compatible key, so switching models costs nothing.
- Pay-as-you-go works with no plan attached, keeping commitment near zero.
- TEE enclaves on featured models, a privacy stance most cheap providers lack.
Cautions #
- Terms changed three times in early 2026 (free tier removed, 5x cap, Base tier model cuts), and a fourth time in October 2026 when the monthly plans stopped being sold, so treat any quota as provisional.
- InfoWorld’s reviewer found performance underwhelming, called the privacy policy ambiguous, and needed a VPN workaround to sign up.
- The March 2026 update cites a global GPU shortage limiting inventory, so capacity depends on miner participation.
- The $10/$20 plans that carried the bundled daily quotas and 6%/10% PAYG discounts are no longer sold, so the discount path is closed to new users and everything bills at PAYG from here.
Pricing #
PAYG is the only path as of 2026-10-29: GLM-5.2 $1.25 in / $3.95 out per 1M, Kimi-K3 $1.85/$9.50, DeepSeek-V3.2 $1.00/$1.00, Qwen3.8-27B $0.039/$1.25, with a cached-input rate listed for every featured model (GLM-5.2 $0.125, Kimi-K3 $0.185) on the ten-model featured table. The Plus $10 and Pro $20 monthly plans are no longer sold: the pricing page carries no plan cards, and the FAQ states that existing plans stay active to the end of the current billing period and then do not renew, with usage billing PAYG from the account balance after that. There is no free tier and no free trial per the FAQ. Private chutes run on an RTX Pro 6000 at $2.25/hour plus a $6.75 one-time deployment fee (3x the hourly rate), up from $1.80/hour and $5.40 at the 2026-10-14 read.
Price history #
| Date | Plan | Change | Source |
|---|---|---|---|
| 2025-07-31 | Subscriptions | Pricing tiers announced, “coming next Monday” (live early August 2025) | Chutes news post, July 31, 2025 |
| 2026-02 | Base / Plus / Pro | Baseline: $3 / $10 / $20 per month tiers existed | Feb 27, 2026 community announcement usage tables |
| 2026-02-27 | Early Access | Free 200-requests/day perk cut off from TEE models, fully retired March 15, 2026 | same announcement |
| 2026-02-27 | All subscriptions | Value capped at 5x PAYG equivalent; GLM-5, Kimi K2.5, Qwen 3.5, MiniMax M2.5 removed from Base | same announcement |
| 2026-09-26 | Base | No longer listed on the pricing page, which shows only Plus $10, Pro $20, and Enterprise | chutes.ai/pricing |
| 2026-10-14 | PAYG (Kimi-K3) | Kimi-K3 listed at $1.85/$9.50 per 1M tokens, down from the $3.00/$15.00 read of 2026-10-11; GLM-5.2, Qwen3-235B, and DeepSeek-V3.2 rates, Plus $10, Pro $20, and private GPU pricing unchanged | |
| 2026-10-29 (observed) | Private chutes | Self-serve TEE GPU rate rose to $2.25/hour with a $6.75 deployment fee (from $1.80/hour and $5.40), keeping the 3x deploy-fee multiple | |
| 2026-10-29 (observed) | All subscriptions | Monthly plans withdrawn from sale: the pricing page carries only PAYG and private chutes, and the FAQ states existing plans run to the end of their billing period without renewal, ending the Plus/Pro quota-and-discount structure for new users |
Compared to #
- Cerebras Code sells raw speed on one coding model at $50/$200; choose it when latency dominates, Chutes when price dominates.
- GLM Coding Plan sells a large fixed quota on one model family; choose it if you will burn the quota, Chutes for variety without commitment.
Bottom line #
Recommended for hobbyists and cost-driven developers who want many open models cheaply and can tolerate churn in terms and capacity. Not for teams needing SLAs, contractual stability, or closed frontier models. Claim to disagree with: after the 5x cap, Plus and Pro make little sense for coders, because PAYG with a small prepaid balance delivers the same upside without a recurring bill.
Changes #
- 2026-09-26 - Created.
- 2026-10-06 - Cleaned the Compared-to and See also link labels (stray leading dashes inside the link text), with the pricing page re-verified unchanged.
- 2026-10-14 - PAYG re-check: Kimi-K3 moved to $1.85/$9.50 per 1M tokens from $3.00/$15.00, with the subscription tiers and the other verified PAYG rates unchanged; price history row appended.
- 2026-10-29 - Rotation re-verification with a subscription retirement on the page: the Plus/Pro monthly plans are no longer sold (the pricing page carries PAYG and private chutes only, existing plans run out their billing period without renewal per the FAQ), the self-serve TEE GPU rate rose to $2.25/hour with a $6.75 deployment fee (from $1.80/$5.40), the featured model table now carries cached-input rates with Kimi-K3 PAYG holding at $1.85/$9.50, and the FAQ adds no free tier and no free trial; every cited URL re-fetched live this run, price history rows appended.
See also #
- Cerebras Code - the speed-first subscription alternative.
- GLM Coding Plan - the fixed-quota rival that wins on coding throughput per dollar.
- OpenRouter rankings - where provider model traffic becomes visible.
- Model provider feature matrix - how access providers compare feature by feature.
References #
https://chutes.ai/pricing - current PAYG model rates with cached-input columns, private GPU pricing at $2.25/hour and a $6.75 deploy fee, and the FAQ stating monthly plans are no longer sold (fetched, HTTP 200, re-verified unchanged 2026-10-11; re-fetched 2026-10-14 with the Kimi-K3 rate at $1.85/$9.50; re-fetched 2026-10-29 with the plans withdrawn from sale, the GPU rate at $2.25/hour, and the no-free-tier FAQ statement).
https://www.infoworld.com/article/4075825/how-to-vibe-code-for-free-or-almost-free.html - critical third-party review: underwhelming performance, ambiguous privacy policy, signup bugs; confirms the $3 entry tier in October 2025 (fetched, HTTP 200).