Sovereign AI at all levels (national, business, personal) has never been more important. Technologies that enable the creation and ownership of frontier intelligence are going to become the new bow wave of this future.

Cracked.

Introducing Decentralized Language Models (DeLM)! DeLM is a multi-agent framework that enables asynchronous, verified & reusable progress! It makes agentic tasks more accurate and significantly cheaper. For example, it achieves 65.7% on SWE-bench Verified using Gemini 3-Flash, a ~10% jump over the best centralized alternatives at less than half the cost. Great work led by @Mao_Yuzhen !

Or…

This is not surprising. It's interesting, however, that the most precious secrets are related to pretraining and distributed training methodologies. Models like Fable are reaching the point they cannot be captured by a single cluster, even the largest like Colossus 2. Finding ways of liquidising training workloads to work over pools of compute, and fit in and around peaking, spiking inference demand is clearly part of the frontier moat.

Many subnets have been doing exactly this for a while now Apex (Subnet 1) is one of the most interesting, with the ability to host multiple problems End state of Apex is: - anyone show's up with a problem - gets translated into an optimization problem - miners (agents) work to solve it until a satisfactory solution is produced - miners open-sourcing their solutions means agents end up collaborating together

I worked in construction for over a decade We are witnessing Elon leveraging his exceptional skill in project execution over competitors. Over 81% of projects end up in delay. These infrastructure delays will cascade into compute. Efficiency of compute is so so critical.

powered exclusively by phenomenal knitwear and conditioner.

what an amazing, wonderful week at @proofoftalk. A few thoughts: - we have come a LONG way in a year. - Teams that were shuttling first dollar in 25 are hitting 8 figures mid 26 (shout out to my brother @0xcarro and the @manifold team, and partnership warlord @MaxScore and the gang) - research subnets with dreams in 25 are publishing actual results - druggable receptors for ADHD meds! Big up @metanova_labs (and @micaelabazo) - lots of bullish new teams (@cacheon_ai Babelbit, Green Compute), and returning first timers from 25 who have made series strides this year ( @markjeffrey hosting the fireside only 13 months after his first in person event at Endgame, after launching a fund. Shout out to @MarkCreaser at DSV) - @Old_Samster combining his increasingly sharp alpha with a new all white angelic look. - apparently @taostats is the new Da Vinci stage (which as a Bittensor oriented stage rocked) - does that mean @mogmachine is the statue of David? - my brother @macrozack on stage proving pitches, and taking the finishing slot of the conference this year with @bitstarterAI was a sweet, full circle moment. -thank you to Paul and the @XVentures_Fund team for another stellar performance. As for us, last year was when we started the @IOTA_SN9 dream. This was the week it felt like coming onto reality. Big love to everyone who’s been a part of it. 2026 is clearly the year.

Just orchestrated a 128 node permissionless decentralized training run, in 5 minutes, for 5 TAO, via @IOTA_SN9 They can do this up to 100B param models. Unbelievable. iota.macrocosmos.ai/dashboar…

ORION LAUNCHES: @macrocosmosai unveils Orion-100B at @proofoftalk on the 1 year anniversary of @IOTA_SN9 launch in this very building. They asked: Can frontier-scale AI training be distributed? On Monday, the team published Orion-100B, an early pretraining run designed to test exactly that. At Proof of Talk co-founders @macrocrux and @WSquires took the audience through the results. Orion-100B represents the largest distributed LLM pretraining run conducted over the open internet to date. The run trained a 100-billion-parameter model architecture across geographically distributed infrastructure. This was not an attempt to produce a finished frontier model. Orion-100B processed approximately 1.1 billion training tokens over a two-day period before being stopped. The objective was simpler, and arguably more important: demonstrate that training at 100B scale is possible without relying on a single, centralized cluster of GPUs. For Macrocosmos, the result represents the culmination of more than a year of work. The version of the story shared on stage was refreshingly unglamorous. The team launched Subnet 9 in 2025 targeting a 15B parameter model and quickly discovered that building distributed training systems in a permissionless environment is harder than it looks. The network struggled. Assumptions broke. The architecture was reworked. So instead of pushing forwards, they scaled backwards. For months, the team trained smaller 1.5B parameter models, running more than 700 experiments in the process. By their own admission, it wasn't particularly exciting. But it allowed them to harden the networking layer, improve fault tolerance, increase throughput and gradually remove bottlenecks from the system. Only then did they begin scaling again. 8B. 18B. Then 100B. The result was Orion. According to the figures presented, the run achieved average model FLOP utilisation of 30.8%, roughly 65% of the speed of equivalent co-located infrastructure, at a third of the cost. More importantly, the learning dynamics remained stable throughout the run, even as the system handled synchronisation and communication across dozens of distributed devices. The technical depth in the room was high. Questions came in on heterogeneous compute stacking, minimum GPU yield thresholds, the reconstructability of sharded weights, architectural constraints on the ResBLM approach. Halfway through, @const_reborn materialised in the audience and started grilling them on the economics and fault tolerance at scale. They took it in elegant stride. The Macrocosmos thesis is that the future of AI training does not necessarily belong to ever-larger datacentres. If distributed systems can become sufficiently efficient, they could unlock vast pools of underutilised compute spread across the world. Today's Orion run was intentionally conservative. The GPUs were distributed, but still professionally provisioned. The next stages of Project Orion will progressively introduce heterogeneous hardware, interruptible spot instances, permissionless participation and eventually consumer-grade devices. Whether distributed training ultimately reshapes the economics of AI remains to be seen. But either way, Macrocosmos has moved the conversation beyond whether distributed training is possible and towards how far it can scale. The presentation also provided useful context for @bitstarterai's newly announced ML Track, which was unveiled by @macrozack at the same roundtable event and is linked by a common theme - how Bittensor attracts and supports the next generation of machine learning teams. Throughout the event, bitstarter.ai outlined its approach to reducing the barriers to entry for researchers looking to build on the network, combining subnet funding, infrastructure support, compute resources, partnerships and incubation. If Macrocosmos' journey demonstrates what is possible once a team is established on Bittensor, bitstarter.ai's ambition is to help more teams make that journey in the first place. For now, however, the spotlight belonged to Orion. After more than 750 experiments, a year of iteration, and a successful 100B-scale demonstration, Macrocosmos has moved distributed training a little further out of the realm of theory and a little closer to reality. And judging by the reaction in the room, plenty of people were paying attention.

We are still so early. Decentralised training has the ability to materially disrupt how inference and training workloads are balanced, and impact both compute utilisation and pricing at a macro scale. We’ll be talking more about our vision for this in the coming weeks. Thanks to @jbrukh for flying the training flag.

The most important thing to understand about Orion is it closes the circle on the thesis for @IOTA_SN9. Not just frontier scale models, but using compute acquired at a fraction of the cost, orchestrated together with MFU approaching frontier performance.

The prime driver of cost in pre training is MFU, or simply put ‘how often are the GPUs actually being used during training’. What if interruptible, distributed paradigms could achieve MFUs comparable or better than centralised? What if…

I will also be there, but apparently my head is too large for this poster. Can’t wait!

I like big nodes and I cannot lie.

First they ignore you, Then they laugh at you, Then they fight you, Then you train bigger models than them.

people call iota a "sleeping giant", but we are very much awake. The activity on X has no correlation to the quality of work we do, which will be considered generational