Subnets / SN65
TAO Private Network SN65
TAO Private Network (subnet 65) is a Bittensor subnet held by 2,137 unique coldkeys. Its liquidity pool holds 6.0k τ, with the alpha token trading at 0.0025 τ. Roughly 2% of tracked holders are in profit — most bought in above today's price. Seller exhaustion is elevated: 89% of recent sellers have emptied their wallets, thinning the supply overhang. Full insider flow, whale tracking and the complete proprietary factor set for TAO Private Network are available to SubnetStats subscribers.
ALPHA PRICE
0.0025 τ
IN TAO
HOLDERS
2,137
UNIQUE COLDKEYS
POOL DEPTH
6.0k τ
TAO IN POOL
7D MOMENTUM
-6.7%
ALPHA vs TAO
Project activity
TPN-004 is in scoring.
Submissions for TPN's 4th competition closed today.
The subnet requested Magistral-Small-2509-Uncensored compressed under a 16GB memory ceiling, scored on IFEval 40%, GSM8K 35%, ARC-Challenge 25%.
We're looking forward to test the model that'll come out.
TPN-004 is open!
Subnet 65 will optimize Magistral Small 2509 Ultra Uncensored from 47GB to a 16GB ceiling.
We are at our 4th competition, and the ceiling has doubled every round, 2GB, 4GB, 8GB, now 16GB.
The scoring changed too, as earlier rounds tested broad knowledge, the 4
TPN products update.
We have built several products around TPN since pivot, and we are currently fine tuning each one of them to be used at scale, by anyone.
TPN’s subnet 65 compresses and optimizes LLMs, producing models that can fit smaller hardware while retaining performanc
TPN-002 is scored.
We've had 21 independent attempts at optimizing Qwen3.5-9B to fit it under a 4GB ceiling, and hold it across five benchmarks at once.
Every model produced by TPN is more sophisticated than the previous one and lays the ground for the model that comes next.
Introducing TPN-003
TPN's third competition, and the largest model yet.
Miners are tasked to take Gemma4-12b-uncensored under an 8GB ceiling, scored across five benchmarks.
A 12B model is roughly 24GB at full precision. Getting it under 8GB enables it to move from a rented GPU
Two weeks on mainnet.
Two weeks ago, True Performance Network went live. Since then, 17 miners have taken Qwen3.5-2B and compressed it down to a 2GB ceiling, producing the first ever TPN optimized model, and proving the mechanism works.
Right now, Subnet 65 is hosting tpn-002
TPN-002 is live.
With Genesis, TPN compressed Qwen3.5-2B to half its size in an open competition.
This time the scope is to compress @Alibaba_Qwen 3.5-9B under a 4GB memory ceiling while retaining capability across five benchmarks.
Genesis measured what survives compression i
Genesis is scored and TPN has its first winners.
Subnet 65 just optimized its first model in an open competition. @Alibaba_Qwen's Qwen3.5-2B went in at 4GB.
The output model, "tpn-001: Ninja over 9000" came out at 1.89GB measured, scoring 0.6487 MMLU and 0.9278 HellaSwag.
Ge
The first round of TPN Genesis just closed.
17 miners committed submissions before the commit/reveal window ended.
A week ago TPN proposed the challenge to take Qwen3.5-2B from 4GB down to 2GB, keeping as much of the model's intelligence intact as possible. Now 17 teams have do
Thanks @taodaily_io for covering TPN Genesis.
This competition's winning model will be the network's first proof that subnet65 works, and will be published on @huggingface & TPN own portal for anyone to use.
Link to Taodaily's article below. https://t.co/NmL5SheoGf
The TPN gateway is live.
Enterprises and devs alike can now submit a model, specify a hardware ceiling or a benchmark floor, and get back a model optimized by TPN to meet it.
Devs use it to fit models onto local hardware; teams use it to cut what inference costs them.
👇 https:
TPN Dev Update
A few things just shipped:
a) TPN's emissions are no longer burning at 100%.
TPN emissions were set to 100% burn during our pivot, since there was no real work yet to reward.
Emissions are now allocated per competition (not at a fixed rate). TPN Genesis is the
Introducing TPN Bench: LLM Benchmarking as a Service
TPN Bench is the infrastructure TPN uses to independently verify every model submitted via a competition, and it's now available as a standalone product for everyone to use.
By uploading a model on Bench, it provisions GPU ca
https://t.co/3fjJwgyGg9
TPN is live on Bittensor mainnet.
We’re excited to share that subnet 65's first competition, TPN Genesis, has officially begun.
We picked Qwen3.5-2B as the target, a compact open-weight model from Alibaba's Qwen family, sitting around 4GB at full precision.
The aim is to compr
TPN is ready for mainnet.
We've spent the last weeks running competitions on testnet, and we fixed what needed fixing between overlapping cycles, benchmarking and miners submissions.
The system is working as per intended and TPN is ready to go live with its first competition th
TPN is ready.
xx/Aug/2026 https://t.co/aSn1QZJmqI
A growing share of AI's future will run locally, offline and privately.
But edge AI (phones, cars, offline devices) has a size vs. intelligence problem, as massive models can't be crammed into tiny chips.
TPN's first mainnet competition is almost here, built to provide smaller
We're currently testing the competition flow, where each competition strengthens the next.
TPN's competitive moat scales with every cycle. As more miners compete, more results are cached, making validation cheaper and faster over time.
This ultimately drives better results at
TPN dev update.
In depth testing on testnet has officially started.
We're running competitions end to end, commit, reveal, scoring, benchmarking, finding edge cases and squashing bugs in real time.
The first competition on mainnet is getting close. https://t.co/2J3K5fmGR9
Nothing about our plan for TPN depends on where we sit today.
We're building Subnet 65 to become the go-to LLM optimization engine, with a product serving a market projected to reach $48B by 2035.
We know what we're building, and we have no doubts about our trajectory. https://
Two types of users come to TPN with two very different problems.
A) "Make the model as small as possible without dropping below our benchmark floor".
B) "Make the model as powerful as possible within our hardware constraints".
What we're building is an engine meant to provide a
Reminder, if you haven't signed up yet, now is a good time.
The litepaper is out and the first miner's competition is next in line.
Sign up to get notified 24 hours before we go live, or register as an enterprise to be first in line to optimize your model.
The TPN litepaper v0.1 is live.
True Performance Network is a decentralized AI model optimization engine on Bittensor, built on the thesis that the most valuable model is the one that can actually run under a given hardware constraint.
Subnet 65 runs competitions between miners
We're building toward the first miner competition on subnet 65.
This event will kick off TPN's optimization engine, bringing a solution to Bittensor for one of AI's most critical problems: How do you preserve frontier level intelligence while fitting into the hardware constraint
SELLERS DRAINED
89%
28D SELLERS EMPTIED OUT
CAPITULATION
+0.58
BOUNCE COMPOSITE (z)
HEAT 7D
0.27×
VOLUME ÷ POOL DEPTH
HOLDERS IN PROFIT
2%
TRACKED WALLETS · ENTRY BELOW PRICE
See who's buying and selling TAO Private Network — every trade classified miner / validator / owner / outsider, plus holder cost basis and the full 23-factor screener.
OPEN THE INSIDER TAPE →