Subnets / SN64
Chutes SN64
Chutes (subnet 64) is a Bittensor subnet held by 13,296 unique coldkeys. Its liquidity pool holds 200.4k τ, with the alpha token trading at 0.0694 τ. Roughly 2% of tracked holders are in profit — most bought in above today's price. Seller exhaustion is elevated: 72% of recent sellers have emptied their wallets, thinning the supply overhang. Full insider flow, whale tracking and the complete proprietary factor set for Chutes are available to SubnetStats subscribers.
ALPHA PRICE
0.0694 τ
IN TAO
HOLDERS
13,296
UNIQUE COLDKEYS
POOL DEPTH
200.4k τ
TAO IN POOL
7D MOMENTUM
-5.9%
ALPHA vs TAO
Project activity
Efficiency-maxxing is the new meta. Find out how Jon is changing the game @ExploitSummit later this month.
Today the Chutes team is happy to introduce our new CEO, Christian Hyland @cjohan247.
Christian has worked closely with us for the past several months, developing new approaches to how we scale and operate in a fast-moving AI market. His contributions have already reshaped how w
Or swap to Chutes.
Three simple changes to get back to work.
https://t.co/N4selsNL0M
Week 16. The rundown:
→ One year of Chutes production traffic is now a research paper: 6.12B requests, 35.8T input tokens, 314,970 users. Co-authored with Harvard and UChicago. Request-level metadata only, nobody's prompts are in it: https://t.co/3tUavjYdK0
→ The paper's findin
https://t.co/axjzWEiufA
NVIDIA's own researchers estimate that 40 to 70% of agent LLM calls could be offloaded to small language models.
Most of an agent's loop is narrow, repeatable work. Routing a request, extracting a field, formatting an output, calling a tool. That is small-model work at small-mod
Security you cannot read is a promise.
Chutes open-sourced its TEE stack (sek8s) and the end-to-end encryption layer that runs on top of it. The architecture guarding your data is code anyone can audit.
Attestation proves a real enclave is running. Open code shows what runs ins
>6 billion requests, trillions of tokens across over 9000 models.
Here's what we learned, in research led by Harvard assistant professor Juncheng Yang @1a1a11a
(thank you to everyone who participated in our research opt-in!) https://t.co/3vJ9UgVeqC
Week 15. The rundown:
→ We said out loud where this is headed: not an AI company, an Intelligence Federation meant to outlast any one operator, including us. Full thread: https://t.co/Sv5FK7w9sv
→ DeepSeek V4 Flash in the spotlight: $0.44/M in, 1M context, in TEE
→ Jon's DGX Spa
https://t.co/KyFTX5VePD
DeepSeek V4 Flash runs on Chutes at $0.44 per million input tokens, $1.32 out. 1M-token context.
Inside a hardware-attested TEE with end-to-end encryption, like every model we serve. Nobody reads your prompts on the way through.
Try here: https://t.co/aCZhC1Gelr https://t.co/U
Wanna get started on Chutes but not sure how?
The starter guide at https://t.co/eVeRj5sWxo
takes you start to finish, account creation to a live API call in one sitting.
Any OpenAI-compatible client works, just swap the base URL to https://t.co/X8XyQ8xgj5. https://t.co/DxeS
We’ve been thinking a lot about this, lately... How do you build something that can last a thousand years?
Almost no company has done it. Most countries haven't either.
Chutes must become a network that outlasts any single company, team, or founder. Including us. https://t.co
We've all heard of binary - zero or one, yes or no.
Ternary is three options: -1, 0, 1.
NVIDIA GPUs have a "sparsity speedup" which means you can nearly double the performance if 2 of every 4 contiguous weights are zero*.
This is the magic - ternary already has natural sp
Meta's Muse Glimmer 30B is live on Chutes at $0.0245 per million input tokens, $0.0978 out. That ties the cheapest tier we serve. Images and text in, 131K context.
It runs inside a hardware-attested TEE with end-to-end encryption, like every model we serve. Nobody reads your pro
Week 14. The rundown:
→ Fez traced where your AI bill actually comes from: chip maker → cloud → model lab → you, every margin stacked. Open models made the priciest layer
Link: https://t.co/cByWXU1oC1
→ Qwen3.8-27B live on Chutes the same weekend the weights dropped. $0.40/M in
https://t.co/KcI6pjjnbh
Your AI bill climbs through the chip maker, the cloud giants, the model labs, then lands on you with every margin stacked on top. That is why it only goes up.
Open models caught up, so the priciest layer is optional now. Make the compute compete underneath and the bill comes dow
Qwen shipped Qwen3.8-27B's weights on Friday. Dense 27B, Apache 2.0, and the model half of X spent the weekend benchmarking. It is already live on Chutes.
Qwen3.8-27B on Chutes: 262K context, text + image input, $0.40 per million input tokens, $3.00 out.
Runs inside a hardware
LTX 2.5 is live on Chutes.
every frame and every sound in this video came off one endpoint. yes, including the music.
sound on.
run it yourself: https://t.co/q2WolTKKuc
text to video, image to video, keyframe interpolation, synced audio. one POST request. https://t.co/H1B0pH
Send a prompt to a hosted AI API and it arrives on someone else's machine as readable text. Staff. Logs. The routing layer. The company that owns the model.
Nothing in the architecture stops any of them from reading it. What stops them is a policy.
Every model served on Chutes
New on Chutes: docuextract.
Feed it a PDF or an image. It hands back clean markdown or schema-guided JSON.
Invoices, contracts, bank statements, scanned forms, research papers. Multi-page documents in a single call.
And like 100 percent of the models served on Chutes, it runs
Week 13. The rundown:
→ Chutes is migrating from the current smart contract mechanism to conviction lock, enforced at the protocol level. More announcements next week
→ Cached prompt tokens: 90% off, up from 50%. Long-context and repeated calls just got a lot cheaper
→ Paralla
https://t.co/ShlQpwgCWj
Someone hands your training run an update. How do you know they did the work you asked for, and not something else?
Jon Durbin's answer in Parallax is that the acceptance test is concrete. Did this update improve loss on the routed expert it was supposed to touch? Then take slic
SELLERS DRAINED
72%
28D SELLERS EMPTIED OUT
CAPITULATION
-0.32
BOUNCE COMPOSITE (z)
HEAT 7D
0.10×
VOLUME ÷ POOL DEPTH
HOLDERS IN PROFIT
2%
TRACKED WALLETS · ENTRY BELOW PRICE
See who's buying and selling Chutes — every trade classified miner / validator / owner / outsider, plus holder cost basis and the full 23-factor screener.
OPEN THE INSIDER TAPE →