Chutes is not cheaper than OpenRouter across the board, because it is a host and OpenRouter often resells it. On 20 October 2026, six of the fourteen models on the Chutes price list were also sold through OpenRouter with Chutes as the host, at exactly the same per-token rate, so going direct saves only OpenRouter’s 5.5% credit fee. Where Chutes is the cheapest host, as on Kimi K2.6 at $1.95 per million output tokens, it wins; on DeepSeek V3.2, DeepSeek V4-Flash and GLM-5.2, OpenRouter reaches hosts that charge a fraction of the Chutes rate.
Same host, same price, and the part most comparisons miss
Chutes and OpenRouter are not two routers competing for the same job. Chutes runs the models itself, on confidential TEE GPUs, and bills per million tokens. OpenRouter runs nothing: it sends your request to one of its 80+ providers, and Chutes is one of them.
So the real comparison is Chutes against every other host OpenRouter can reach for the same model, plus OpenRouter’s fee on top.
Six models where OpenRouter sells Chutes at the Chutes rate
The Chutes pricing page lists fourteen featured models. Six of them also appear on OpenRouter with a Chutes endpoint, and on all six the routed price matches the direct price to the cent:
- Kimi K2.6: $0.45 input, $1.95 output on both sides
- Qwen3.6-27B: $0.25 input, $2.00 output
- Qwen3.8-27B: $0.039 input, $1.25 output
- GLM-5.1: $0.98 input, $3.08 output
- Gemma 4 31B: $0.12 input, $0.37 output
- Kimi K3: $1.85 input, $9.50 output
Cached input follows the same pattern. Chutes bills cache reads at a tenth of the input rate on every model of its API listing, and the OpenRouter endpoints carry the same figure, for example $0.045 on Kimi K2.6. This is the pass-through I measured on the OpenRouter pricing breakdown, applied to a smaller host.
What the 5.5% credit fee does to those six
OpenRouter never marks up tokens. Its money comes from a fee on the credits you buy: 5.5% on the Standard plan, 8% on Business, according to the OpenRouter pricing page. Spread over the tokens, that turns Kimi K2.6 output from $1.95 into an effective $2.06 per million on Standard.
On a model Chutes hosts for OpenRouter, the only saving from going direct is the 5.5% credit fee: $5.50 on every $100 of credits.
On those six models, my verdict is simple. If Chutes is the host you want, buy from Chutes. You pay the same per token and skip the fee.
Where Chutes is the cheapest host
The table compares the Chutes price with the next-cheapest endpoint on OpenRouter for the same model, ranked by output price, with temporary promotions left out. Prices are in dollars per million tokens, input then output.
| Model | Chutes against the next host on OpenRouter |
|---|---|
| Qwen3.6-27B | Chutes $0.25 / $2.00; Reka $0.30 / $3.00 |
| Kimi K2.6 | Chutes $0.45 / $1.95; DigitalOcean $0.57 / $2.40 |
| GLM-5.1 | Chutes $0.98 / $3.08; SiliconFlow $1.19 / $3.74; Z.AI $1.40 / $4.40 |
Kimi K2.6 and Qwen3.6-27B
These two are the clearest Chutes wins. On Qwen3.6-27B, Chutes is the cheapest endpoint on both input and output, and a third cheaper on output than Reka. On Kimi K2.6, Inceptron undercuts Chutes on input ($0.441 against $0.45) but charges $3.20 for output, so any workload that writes more than it reads lands on Chutes.
Even routed through OpenRouter with the fee, Kimi K2.6 on its Chutes endpoint comes to about $2.06 per million output tokens, against $2.53 for DigitalOcean with the same fee. OpenRouter’s default routing favours low-priced hosts, so Chutes is already in the running on Kimi K2.6 without any setting.
GLM-5.1, once the promotions are set aside
StreamLake lists GLM-5.1 at $0.966 / $3.036, a hair under Chutes, but with a 31% discount flagged on the endpoint. Discounts on OpenRouter endpoints come and go, so I would not build a budget on that one. At list prices, Chutes beats every host including Z.AI, the lab that trains the model.
Where OpenRouter finds a cheaper host than Chutes
The other eight Chutes models have no Chutes endpoint on OpenRouter. On most of them, OpenRouter reaches hosts that charge far less, and this is where the “cheaper router” reputation breaks.
| Model | Chutes direct against the cheapest host on OpenRouter |
|---|---|
| DeepSeek V4-Flash (0731) | Chutes $0.44 / $1.32; OpenInference $0.0121 / $0.12; DeepInfra FP8 $0.06 / $0.18 |
| DeepSeek V3.2 | Chutes $1.00 / $1.00; AtlasCloud FP8 $0.26 / $0.38 |
| GLM-5.2 | Chutes $1.25 / $3.95; DigitalOcean $0.70 / $2.20 |
| Qwen3-32B | Chutes $0.104 / $0.416; DeepInfra FP8 $0.08 / $0.28 |
DeepSeek V3.2 and V4-Flash
DeepSeek is where Chutes is weakest. DeepInfra serves V4-Flash at the same FP8 quantization as Chutes for $0.18 per million output tokens; with the 5.5% fee that is about $0.19, roughly seven times below the Chutes rate of $1.32. On V3.2, AtlasCloud’s FP8 endpoint at $0.38 output, about $0.40 with the fee, is two and a half times cheaper than Chutes at a flat $1.00.
DeepSeek V4-Flash (0731): $1.32 per million output tokens on Chutes, $0.18 at DeepInfra through OpenRouter, same FP8 quantization.
GLM-5.2, cheaper than the lab and still not the cheapest
Chutes charges $1.25 / $3.95 for GLM-5.2, below the $1.40 / $4.40 Z.AI asks on its own endpoint. That looks like a deal until you see DigitalOcean at $0.70 / $2.20 and CoreWeave at $0.62 / $2.42 on the same OpenRouter page. I covered this pattern on the OpenRouter model comparison: one model name hides a wide spread of hosts.
Even among the six models Chutes hosts for OpenRouter, three have a cheaper neighbour: Qwen3.8-27B at Tenstorrent ($0.02 / $0.50), Gemma 4 31B at CoreWeave ($0.10 / $0.34) and Kimi K3 at Sail Research ($1.20 / $9.00).
Chutes Plus and Pro subscriptions, read closely
Chutes is pay-as-you-go by default, with no minimum. Two optional monthly plans add a daily request quota and a per-token discount, something OpenRouter does not sell.
Chutes announced the discount that applies past the cap in this post; the Base tier it names is no longer on the pricing page.
With the recent subscription updates, we wanted to make sure subscribers who go beyond their cap aren’t just dropped into standard rates.
So we’re adding automatic PAYGO discounts for all subscription tiers:
→ Base: 3% off
→ Plus: 6% off
→ Pro: 10% off
When you hit your sub https://t.co/RXLEBZSuk8@chutes_ai
https://x.com/chutes_ai/status/2028523782143361453
The caps that bite before the request count
Plus costs $10 a month for 2,000 requests a day, Pro $20 for 5,000. The request count is the headline, but each plan also carries a monthly usage cap of five times its price, measured in pay-as-you-go dollars, and a spend cap per rolling four-hour window.
- Plus: $50 monthly usage cap, $4.17 per four hours, 6% off pay-as-you-go beyond the quota
- Pro: $100 monthly usage cap, $8.33 per four hours, 10% off pay-as-you-go beyond the quota
Chutes says so itself: the daily request limit does not guarantee full usage, because the monthly or burst cap may be reached first. On Kimi K3 at $9.50 per million output tokens, $4.17 buys about 440,000 output tokens, so a long agent session can reach the Plus burst cap long before the daily request count.
Why an app cannot run on a plan
The pricing FAQ is explicit: subscriptions are for casual, direct use, and automated or high-volume workloads must use pay-as-you-go. For anything you ship to users, the comparison is pay-as-you-go Chutes against OpenRouter, which is the one the tables above make.
What each one gives you besides the price
When the price gap is small, these differences decide.
- Fallback: OpenRouter retries other hosts when one fails and does not bill the failed attempt. A direct Chutes key has one host.
- Catalog: OpenRouter lists 500+ models on paid plans, including closed models from the big labs. Chutes features fourteen on its price page, with more in its app catalog.
- Free usage: Chutes has no free tier. OpenRouter’s free models allow 50 requests a day, 1,000 once you have bought $10 of credits, and Gemma 4 31B is among them.
- Limits per endpoint: on Kimi K2.6, the Chutes endpoint on OpenRouter caps a completion at 65,535 tokens and only accepts tool_choice set to auto, where Inceptron allows 235,929 tokens and forced tool calls.
- Confidential compute: every featured Chutes model runs on TEE hardware, which matters if your prompts carry customer data.
- Payment: Chutes takes cards, crypto and TAO, the Bittensor token. OpenRouter takes cards, AliPay and USDC.
For the wider field, my list of OpenRouter alternatives covers the gateways that charge a subscription instead of a fee.
Which one to put in front of your app
My verdict depends on the model, not on the brand.
- You run Kimi K2.6, Qwen3.6-27B or GLM-5.1: go direct to Chutes. It is the cheapest host and you skip the 5.5% fee.
- You run DeepSeek, GLM-5.2 or Qwen3-32B: use OpenRouter and pin the cheap host with the provider order field, or set sort to price. The fee is small next to the gap.
- You mix several open models and need failover: stay on OpenRouter. Its default provider routing weights stable hosts by one over the square of their price, so the cheapest host, Chutes included, gets the largest share of requests.
- You chat with open models by hand every day: Chutes Plus or Pro, as long as your sessions stay inside the four-hour cap.
Frequently asked questions
Is Chutes available on OpenRouter?
Yes, as a provider. On 20 October 2026 OpenRouter listed Chutes endpoints for six models, including Kimi K2.6, Qwen3.6-27B and GLM-5.1, at the same per-token price as the Chutes site.
Does Chutes have a free tier?
No. Chutes states it has no free tier and asks you to top up a balance, with no minimum. OpenRouter keeps a set of free models with daily request limits.