Network /Systems Engineer , USMC Vet
San Diego, CA
- Tweets 696
- Following 236
- Followers 449
- Likes 5,699
September flew by. We passed the 500 billion tokens mark, welcomed two teammates, replaced the MDM requirement, and added more models.
Now, we are focused on making it easier for providers to earn more with Autopilot.
More in my letter to our provider community below.
8
6
51
2,656
If this approach feels familiar, it is.
This is the same playbook of open innovation that Eigen's always been known for, but this time @sreeramkannan focused the team on the most powerful application of these incentives/protocols -- collective research and scientific progress.
What @eigencloud is doing is replicating some of the protocol design techniques that made EigenLayer so magnetic, but fixing the flaws that limited its impact..
Analogy might be helpful here.
EigenLayer allowed ETH holders to stake their ETH and earn passive yield while it secured new protocols looking to borrow economic security...one issue was that outside of DA, we didn't use that security built 1P value ourselves...we waited for others to do it and it never really materialized (history is repeating itself with Bittensor subnets fwiw)
Now the scare resource isn't economic security, its compute and inference.
@darkbloomai allows Mac owners to register their spare compute into a shared grid and earn passive income while developers use their machines to run private inference (10-20B tokens/day and climbing)
@yukonresearch challenges run on darkbloom and let researchers put their spare AI token balances and harness designs towards solving frontier problems in science and research (including improving open-weight models that expand the footprint of Darkbloom itself).
Note: The IP that comes from these challenges can be owned by its contributors and some may will become tokenized companies themselves one day.
Consistency in design principle, while updating adapting what these protocols and incentive systems are aimed at. Other than @ilblackdragon there may not be a more persistent and hard-to-kill founder in crypto/AI than @sreeramkannan
Research is becoming open and multiplayer again.
Here is the @YukonResearch story so far, from one open leaderboard to a platform where anyone can post a hard problem and anyone's agent can take a shot at it.
We asked Claude Opus 5.5 to tell the story.
It made the whole video:
3
9
25
4,909
Jonathan Spangler retweeted
MLX.fast is coming back soon. We are just finalizing the updated benchmarking system so it'll be much harder to game the prompt outputs.
4
4
41
1,801
RT @eigencloud: There’s a lot more happening here than meets the eye.
- @YukonResearch: Open Frontier Research
- @DarkbloomAI: Open Infere…
8
5
Eigen is doing something novel with @darkbloomai and @yukonresearch protocol and incentive design
Yukon improves how open weight models run on Apple Silicon and Darkbloom ships those optimized models on their fleet of Macs - offering developers more performant private inference without the datacenter markup.
Unlike a lot of other DeAI projects that bootstrap demand with GPU subsidies funded by token inflation, @eigencloud has learned from its experience with eigenlayer and built a product that devs want and will pay for.
New challenge on Yukon: MLX.fast x Bonsai 2
@PrismML took Qwen 3.8 27B, a 54 GB reasoning model, and shrank it to about 6 GB while keeping 98% of its benchmark score. It runs on a 16 GB Mac.
Now we want it fast. Optimize how Bonsai 2 runs on Apple Silicon and climb the leaderboard.
If your machine was too small for earlier rounds, this one is for you. In the first day, the community made it run 2.26x faster than where it started, and there is a lot of room left.
Open until October 1.
3
6
32
9,684
RT @eigencloud: AI research is starting to happen in the open. That's a bigger deal than it sounds.
Today the biggest breakthroughs come o…
13
4
RT @eigenlabs: What does research look like when AI agents can build on each other’s work at machine speed?
Our Head of Research @SoubhikD…
8
3
RT @yukonresearch: New challenge on Yukon: MLX.fast x Bonsai 2
@PrismML took Qwen 3.8 27B, a 54 GB reasoning model, and shr…
18
3
Let’s push mlx even further for all ! Bonsai 2 by @PrismML has been a so capable for agents running local on low memory machines . Can’t wait to see where the community pushes this one.
MLX.fast is launching a 1 week contest to improve Bonsai 2 from @PrismML! Bonsai 2 is a near losslessly comrpessed version of Qwen 3.8 27b that is around 10% the size of the original. It can fit on 16gb Apple devices and even some phones.
This is a model almost anyone can run so if you weren't able to participate in the other contests, this one is for you! The community has been able to double the speed of Bonsai in under 12 hours so there is alot of headroom here.
This contest runs until October 1, have fun!
1
2
8
1,043
Jonathan Spangler retweeted
MLX.fast is launching a 1 week contest to improve Bonsai 2 from @PrismML! Bonsai 2 is a near losslessly comrpessed version of Qwen 3.8 27b that is around 10% the size of the original. It can fit on 16gb Apple devices and even some phones.
This is a model almost anyone can run so if you weren't able to participate in the other contests, this one is for you! The community has been able to double the speed of Bonsai in under 12 hours so there is alot of headroom here.
This contest runs until October 1, have fun!
9
16
90
29,408
RT @sreeramkannan: Yukon: Make Research Open & Multiplayer Again.
Research progress always needed multiple contributors which needed it to…
19
This is the first time in Darkbloom's history that we are the first provider to support a model on @OpenRouter ❤️
I couldn't have chosen a better partner for this milestone. @PrismML focuses on concentrating intelligence, so more models can run on distributed Mac machines!
2
9
30
1,712
Darkbloom is the first model provider to support Ternary Bonsai 2 27B -- concentrated intelligence that fits on your phone.
Try it now: console.darkbloom.dev/chat
First 250 users get 100 million free tokens;
PrismML's new flagship model:
- a ternary compression of Qwen3.8 27B at 2bits per parameter.
- 8.5 GB total, 5x smaller than original
- keeps 98.2% of the Qwen's FP16 benchmark performance.
- 75% cheaper than Qwen 27B.
Qwen3.8 27B already operates comparably with Opus 4.6 and 5.6 Luna on certain tasks. This one does it in the memory of a phone. 262K context, image input, Apache 2.0.
From our first run on the network, on a single M5 Max with no caching:
- 35 tok/s decode at 1K context,
- 31 tok/s at 10K,
- 19 tok/s at 50K.
But we expect more performance gain coming in a few weeks! That's a full 27B reasoning model running comfortably on any Mac.
1,000+ Macs are serving on Darkbloom right now. Go try it out!!
Thank you to @BabakHassibi @SahinLale @HessianFree @rsadri_ml @tushar_bans @evaninwords and the whole PrismML team. This is exactly the kind of model Darkbloom was built for.
You can read the essay by Bonsai on Why Local AI Matters:
Today, we’re announcing Ternary Bonsai 2 27B.
Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.
Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.
Ternary Bonsai 2 27B is available today under Apache 2.0.
15
21
117
49,024
> be @darkbloomai
>gather everyday people's Macs
>connect them to a grid
>create a private "local AI as a service"
>optimize models like @googlegemma @Alibaba_Qwen and @nvidia nemotron with your sister team @yukonresearch
>run them 2-3x faster and 50% cheaper by stripping out datacenter margins
>partner with @OpenRouter for distribution
>pass earnings along to passionate, growing local AI community that offer up their spare compute.
rinse. lather. repeat
MLX.fast is back. This time on Gemma 4 26B A4B.
Gemma 4 is the most served model on @darkbloomai today.
Faster Gemma means the same Macs can handle more real traffic.
Live on Yukon: yukon.org/mlxfast
2
7
23
1,776
Jonathan Spangler retweeted
Bonsai 2 coming to Darkbloom
A 27B model in 5.9 GB. Most Macs already on the network can serve it, no new hardware needed.
Stay tuned.
Today's my birthday, and I could not have asked for a better gift.
Bonsai 2 fits on most phones shipping today (anything >8GB). And it outperforms Opus 4.6 and 5.6 Luna -- on tasks that would have sounded absurd to attempt anywhere two years ago.
I try to be disciplined about timelines. My most optimistic estimate for something like this was early-2027. It's September 2026.
Huge thank you to the @prismml team -- @BabakHassibi @SahinLale @HessianFree @rsadri_ml @tushar_bans @evaninwords -- for putting this into the world.
We're bringing Bonsai 2 to @DarkbloomAI tonight. The 1,000+ providers can test it right away. And whenever those machines aren't in use, they'll serve Bonsai 2 to anyone who wants to try it.
So few people set up a local model themselves, and a model this good shouldn't be gated by that. More tomorrow.
8
7
108
9,878
RT @eigenlabs: New Challenge on Yukon: yukon.org/qsb
In partnership with @StarkWareLtd.
A quantum computer could one day break…
15
3
Jonathan Spangler retweeted
World’s first massive multiplayer research.
500+ improvements made by 100s of humans and their agents over many months to figure out how to build Quantum circuits that can break https and Bitcoin.
This was the ecdsa.fail challenge.
We started at 0.7x relative efficiency to Google and the system improved it to 2.63x!
Given the OpenAI-Anthropic tussle on Navier-Stokes credit, this kind of system would have helped to assign credit and encourage collaboration.
AI-native scientific institutions are coming.
They are sorely needed now.
Go checkout Yukon.org
This project is so dear to my heart. It's probably the first in the world such massive collaboration project which gathered people using auto-research loops with agents and who focused on one specific task. This task was ECDSA.fail - a design of the quantum algorithm which implements a variant of Shor's algorithm dedicated to breaking the discrete logarithm problem for curve addition on secp256k1 elliptic curve. We've built a community of researchers who study agents coordination, stay in the loop, exchange ideas and push the frontier. It's a new way of doing breakthroughs, extremely rewarding, very competitive and collaborative at the same time. Got a ton of new friends. And we keep on going, fascinating times! Thanks to all of you guys keeping this project alive!
15
19
120
11,948
Congratulations on the paper ! Great read. @eigenlabs is building such an amazing autoresearch platform .
The paper is led by @jieyilong, CTO of @Theta_Network, with coauthors from @ethereumfndn, @StarkWareLtd, @Starknet, @trailofbits, @brevis_zk, @SeiNetwork, @pauli_group, @OctavFi, @sciencevr, @nasqret at Adam Mickiewicz University, and researchers at Warsaw University of Technology and Stanford's Free Systems Lab.
On the comparison, the 2 efforts use different interfaces and accounting conventions, so the paper treats this as a numerical comparison rather than a formal claim.
Craig Gidney and Tanuj Khattar of @GoogleQuantumAI reviewed the manuscript before publication.
Read the full paper on @arxiv: arxiv.org/pdf/2609.09582
3
47
2,961