Lightning Fast Uncensored Inference Subnet 46
Bittensor
- Tweets 21
- Following 9
- Followers 124
- Likes 31
Pinned Tweet
The future of intelligence is fast,
some would say Instant.
Our first model is GLM 5.3 (Flash). Validation code is live and Miner code is hours away.
Launching on Subnet 46.
1
2
13
2,168
Instant Inference retweeted
Podcast coming out tomorrow, with @thinkinstant
You're not going to want to miss this one.
1
1
8
1,253
Start focused. Prove the speed. Scale the usage. That’s it.
We’re building accelerated inference with verification that doesn’t require a TEE.
Once validated in production, we will embed Instant into the apps people already use as well as our own.
⚡️🧠
You can disagree, but the product builders I respect keep coming to the same conclusion:
In-app ads will fund a large share of AI inference.
For @thinkinstant, the path is clear.
Phase one: serve 1–2 models with a clear advantage in inference speed and performance.
Phase two: bring that inference into products people already use, while building our own application.
Better products earn distribution. Give users a better experience, faster, at a lower cost, and you give them a reason to switch.
Here’s what excites me most.
We can build revenue beyond the inference API. Applications built on our subnet can direct 100% of their revenue to buybacks and burns, helping us scale burns beyond miner emissions.
The goal is a deflationary alpha token: more tokens burned than emitted.
We see two paths: sell premium inference at 2.5x what we pay miners, or build additional revenue streams on top of the subnet to fund a total of 2.5x miner burns.
Inference can cover miners while our additional endeavors funded by owner emissions can cover itself and stakers.
Emissions from the Bittensor network will ideally be used to increase our price floor aka mcap:LP ratio.
But first, we have to launch and prove the foundation works.
That means verifying inference without a trusted execution environment (TEE). Removing that requirement can give miners more hardware flexibility and developers a better cost and speed profile.
Niki is finishing the final tests of the miner code. I’m focused on growth.
Prove the inference. Earn the usage. Scale the revenue.
Accelerate. ⚡️🧠
2
8
554
Instant Inference retweeted
You can disagree, but the product builders I respect keep coming to the same conclusion:
In-app ads will fund a large share of AI inference.
For @thinkinstant, the path is clear.
Phase one: serve 1–2 models with a clear advantage in inference speed and performance.
Phase two: bring that inference into products people already use, while building our own application.
Better products earn distribution. Give users a better experience, faster, at a lower cost, and you give them a reason to switch.
Here’s what excites me most.
We can build revenue beyond the inference API. Applications built on our subnet can direct 100% of their revenue to buybacks and burns, helping us scale burns beyond miner emissions.
The goal is a deflationary alpha token: more tokens burned than emitted.
We see two paths: sell premium inference at 2.5x what we pay miners, or build additional revenue streams on top of the subnet to fund a total of 2.5x miner burns.
Inference can cover miners while our additional endeavors funded by owner emissions can cover itself and stakers.
Emissions from the Bittensor network will ideally be used to increase our price floor aka mcap:LP ratio.
But first, we have to launch and prove the foundation works.
That means verifying inference without a trusted execution environment (TEE). Removing that requirement can give miners more hardware flexibility and developers a better cost and speed profile.
Niki is finishing the final tests of the miner code. I’m focused on growth.
Prove the inference. Earn the usage. Scale the revenue.
Accelerate. ⚡️🧠
3
6
18
1,502
Fastest and cheapest inference will outperform the costly and limiting frontier labs.
Uncensored > Private
⚡️🧠
If open source wins, AI value accrual goes to whoever can SERVE INFERENCE the fastest, cheapest.
Bitcoin proved that mining amasses compute BETTER and CHEAPER than ANYTHING ELSE.
Bittensor weaponizes this - and wins.
Cheapest inference on earth:
@engyai
@chutes_ai
@say_gm_
@lium_io
@TargonCompute
@actualinc
All Bittensor subnets. $TAO
1
11
683
Instant Inference retweeted
JUST IN: BlackRock says AI compute could be tokenized in the future.
623
605
7,086
1,076,594
Instant Inference retweeted
Miner code first, then accelerated GLM 5.3 (flash) API keys.
⚡️🧠 @thinkinstant
2
1
13
554
Instant Inference retweeted
Regarding the recent claims about subnet 46.
Instant is launching on Subnet 46.
We will let the product speak for us, not a past employee.
A message from the founder of subnet 46, Instant @sebyrubino.
We will let the product speak for itself.
Validator code went live last week. This week we expect live miner code.
Onwards!
1
2
15
2,016
Instant Inference retweeted
A message from the founder of subnet 46, Instant @sebyrubino.
We will let the product speak for itself.
Validator code went live last week. This week we expect live miner code.
Onwards!
2
6
21
3,269
A message from the founder of subnet 46, Instant @sebyrubino.
We will let the product speak for itself.
Validator code went live last week. This week we expect live miner code.
Onwards!
2
6
21
3,269
Instant Inference retweeted
This is a slanderous claim by a past employee who has not contributed for over a month.
Please follow the account that @sebyrubino and the team are NOT locked out of.
>> @thinkinstant <<
Onwards.
2
6
689
Instant Inference retweeted
Please follow this account.
A previous employee who has not contributed in over a month has locked us out of our account.
No changes just clarification.
Follow @thinkinstant for updates on Subnet 46.
A false claim by a past employee has been made against subnet 46 “Instant”.
We will be launching instant and providing updates here.
Thank you for your attention on this very unfortunate matter.
3
2
14
1,950