I cannot overstate how excited I am to share this with the world. 16B trained over the internet by 4090/5090s with help from @green_compute_ and @TargonCompute. MFUs upwards of 40%, with compute all the way from the USA to Hong Kong. @IOTA_SN9 iota.macrocosmos.ai/dashboar…

so dope, trained a 16B model with on what many would call old gpu's but it worked!

Two machine learning teams. One #Bittensor subnet slot. You decide who gets it. At pitchtensor, hosted by @bitstarterAI, two aspiring teams pitch to @macrocrux, @KeithSingery, @maxsebti, @EvanMalanga + @willobrien - then you choose the winner in a live crowdfund, online + in the room. Live from the Exploit stage: luma.com/exploitsummit26

To understand the real significance of this work you need to understand one thing: coordinated heterogeneous compute is the compute fabric of the future. One that will be as ubiquitous as energy out of your wall, but for flops.

Bittensor Subnet 25 @urnetwork is simultaneously launching on main with a truly groundbreaking product live and running a sim on @Apex_SN1 to optimize network performance $TAO Founder @xcolwell is up there with the top developers in Bittensor

The future of training, synthetic data generation, or inference will live on a distributed network of compute. The future is liquid, we are are building it. iota.macrocosmos.ai/

Two ML teams competing. Five judges who know what it takes to build on Bittensor. Meet the pitchtensor panel at @ExploitSummit: → @macrocrux - CTO, Macrocosmos → @KeithSingery - Chief Evangelist, Tao.com → @kenmiyachi - CEO, Bitmind → @mxmsbt - CEO, Score → @EvanMalanga - CRO, Yuma They’ll ask the tough questions. Then it’s your turn. Hear the ideas, weigh the answers, and back the team you think should build on #Bittensor. 29 September - on the OTF stage: luma.com/exploitsummit26

you idiots are smart enough to lie to yourself long enough to think you have it figured out but in reality you just cooked yourself and didn't even know it

Translation: our gross margins are getting competed down to 0 by open source models and our capex burn rate is too high. Let’s maintain our margins with regulatory capture, ban open source models, and slow down the capex arms race. All with a virtue signaling cherry on top.

Nearly a year of Code Arena: WebDev progress compressed into 15 seconds. Each line follows the highest-scoring model from top labs over time, showing the pace of improvement across the ecosystem. In just the last year, the model in the leading spot increased score by +340 pts and the number of frontier labs competing for the top spot expanded from 6 to 10. @AnthropicAI has dominated throughout the year. Although standout releases have jumped to the top spot, most notably the Chinese open-source model Kimi K3 from @Kimi_Moonshot in July. Today, GPT-6 Astra by @OpenAI leads with 1796 pts, followed by @claudeai Fable 5.1 at 1764 pts. The next closest lab is 103 pts away, @Alibaba_Qwen with 1685 pts. Code Arena: WebDev ranks models through head-to-head user preference on real front-end web development tasks. These votes drive the leaderboard that is tracking the frontier.

At @proofoftalk, we revealed Project Orion, on @IOTA_SN9 and asked a simple question: can distributed AI training actually compete commercially? We've spent the whole summer answering it and at Exploit, we will show what we found. luma.com/exploitsummit26

🚨Excited to share that I’ve joined @btlabs_ai as Head of Americas, Business Development. Enterprise demand for open-source tech is getting serious. The next bottleneck is the distribution of its services and solutions. Let’s build together. 💪🏽

i’d rather fail than not try

Claims is live on mainnet. Miners now compete to turn papers into structured claims and source evidence. Validators score the work. The best work earns the weight. Each round builds the canonical graph of science. 450 testnet papers led here. Day 0 starts now.⏱️

heterogeneous TaH is on the horizon! iota.macrocosmos.ai/dashboar… keep track of the Orion progress here!! I’m personally super excited for the millions who have a 3080-5090 grade rig, who will be able to join the biggest decentralized training network SUPER SOON! @IOTA_SN9 keep cooking

🚨 $TAO SN9 @IOTA_SN9 On Monday alone added: 8x L40S from Tokyo 16x A100 from Calgary 14x A6000 from Las Vegas 8x A100 from Kansas 4x A100 from Helsinki RTX 4090s, RTX 5090s, A100s, A6000s, L40S, all mixed together, all contributing. Top node locations right now are Barrow-in-Furness UK (76), Norwich UK (64), Hong Kong (24), Las Vegas (14), with more spread across Houston, Denver, Calgary, Helsinki and more...... And that mixed hardware is now helping train Orion-16B, a 16.2B parameter model running across 226 active nodes, 12 locations and 3 continents. Live numbers: 130.18B tokens processed 97.1K tokens/sec 21.38% MFU Loss down to 3.1624 The important part is not the model. It is the idea SN9 is testing: Does large-scale training really need one perfect cluster of identical GPUs? IOTA, no, no, no.... If a device can hold roughly 10% of the model weights, it can contribute. Instead of forcing the world to build around a few premium datacenters, IOTA will make the compute already spread across the world useful for training. And unlike a lab, the network is visible: 260 miner nodes 11,413 routing paths individual hotkeys live throughput public leaderboard If this keeps scaling, SN9 is not just building decentralized compute. It is attacking one of AI training’s biggest problems and its working. The Supercomputer subnet. $TAO | SN9 | DYOR.

WE ARE LIVE 🚀 Today marks a special occasion in the bittensor world as we unvail the largest new permanent Gpu cluster to hit the ecosystem. 240 Rtx 5090s are now live specifically deployed to serve commercial clients wanting some longer term compute they can trust to be there in 6 , 12 months and beyond - ofcourse all powered by 100% Green energy 🥬 Head on over to green-compute.com are get instant on demand acess with discounts for people paying in alpha . For private clusters and bare metal acess please contact our sales team who will be happy to assist 💪

i love this team

Orion-16B is now training across 3 continents using 225 GPUs. On Monday we added 50 new GPUs: + 8xL40S from Tokyo + 16xA100 from Calgary + 14xA6000 from Las Vegas + 8xA100 from Kansas + 4xA100 from Helsinki Our system does not care which GPUs it is or where it is. If it can hold 10% of the model weights, we can train on it. When your system is built like @IOTA_SN9 there is no GPU scarcity. iota.macrocosmos.ai/dashboar…

Our first robotics competition landed last week. It went well enough that we didn't wait for what was next. This one was conceived on Monday and built in a couple of hours — a whole series of events, each one a different test of what a robot can learn to do: run, jump, climb, balance, and more. Same idea as before, scaled up: agents train models to control a robot through each event, and what they learn lives in the model's weights, behaviour that can, in principle, move to a real machine. One competition was a test, doing a follow up this fast is the point: on Apex, a new challenge goes from idea to live in hours, not months. The competition is open now! apex.macrocosmos.ai/competit…

Today, I am very proud to be launching @urnetwork (SN25) with dear friends Brien Colwell and Keith Singery. Three months ago, in Stockholm, Brien and I came together to discuss the necessity of a free and open internet for all, a mission which Brien has singularly dedicated himself to for the last two years. We start our journey today