With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
Apex is a Bittensor subnet that’s pushing the boundaries of agentic reasoning by incentivizing miners to build workflows that actively reduce hallucinations and improve the inference quality of large language models. What’s compelling about their approach is how validators host an ongoing competitive arena—where LLMs are equipped with tools and function calls—and only the most capable earn TAO emissions.
What sets them apart is the dynamic and asymmetrical nature of their evaluations, which not only power miner rankings but also produce a high-volume, high-quality dataset. We’re talking millions of tokens per day, generated from real-world agentic tasks. That output becomes the training fuel for fine-tuning in Subnet 37, making Apex a key engine for both performance benchmarking and model improvement across the network.
Apex is a Bittensor subnet that’s pushing the boundaries of agentic reasoning by incentivizing miners to build workflows that actively reduce hallucinations and improve the inference quality of large language models. What’s compelling about their approach is how validators host an ongoing competitive arena—where LLMs are equipped with tools and function calls—and only the most capable earn TAO emissions.
What sets them apart is the dynamic and asymmetrical nature of their evaluations, which not only power miner rankings but also produce a high-volume, high-quality dataset. We’re talking millions of tokens per day, generated from real-world agentic tasks. That output becomes the training fuel for fine-tuning in Subnet 37, making Apex a key engine for both performance benchmarking and model improvement across the network.
Apex (Subnet 1) is where Bittensor’s most competitive, cutting-edge research in agentic reasoning unfolds. They’re building the kind of open-source intelligence that can rival and at times outperform centralized models on task-specific functions. This subnet isn’t just part of the network; it’s the crucible where innovation meets performance, shaping the trajectory of decentralized AI.
Leading Bittensor’s Race for Intelligence
Bittensor’s long-term viability depends on continuous breakthroughs that rival proprietary AI systems. Apex plays a central role in this mission. It’s the protocol’s flagship arena — the environment where validators and miners go head-to-head to build more capable agents, and where mechanism design is constantly being refined.
Since its rebuild in early 2024, Apex has evolved to incorporate advanced inference strategies like response ensembling and agentic workflows, adapting quickly to the demands of fast-moving LLM research. They’ve positioned themselves as the backbone for testing and scaling the very systems that will define decentralized AI’s future.
Beyond Centralized Benchmarks
What makes Apex so compelling is that they aren’t just chasing benchmarks; they’re surpassing them. By orchestrating workflows that combine LLMs with agentic reasoning, Apex has shown that open systems can outperform closed models at targeted tasks. This isn’t theoretical; it’s already being demonstrated across real-world use cases.
How Apex Incentivizes Intelligence
1. Transforming Raw Intelligence into User Products
At its core, Apex turns high-quality outputs into digital commodities. Whether it’s through chat interfaces, APIs, or enterprise tooling, users tap into a constantly evolving hivemind — one that’s trained and refined by competition, not central authority. Right now, they’re already powering four distinct products across the network.
2. Expanding the Incentive Landscape
They’ve pioneered mechanisms that reward high-quality, human-like responses — not just the ability to game synthetic prompts. By incorporating organic queries into the validation layer, Apex pushes miners to generate answers that are genuinely useful, grounded, and context-aware.
3. Fueling the Broader Bittensor Ecosystem
Apex isn’t operating in a vacuum. It’s deeply integrated with Subnet 9 for pretraining, Subnet 13 for data structuring and sentiment analysis, and Subnet 37 for fine-tuning. Together, they form a full intelligence pipeline — one that continuously creates, tests, and evolves open-source AI.
The Flagship of Open-Source AI
As the first subnet on Bittensor, Apex continues to be the standard-bearer, not just for its legacy but for what’s next. With each iteration, they’re not just refining competitive intelligence — they’re defining what decentralized AI can become.
Apex (Subnet 1) is where Bittensor’s most competitive, cutting-edge research in agentic reasoning unfolds. They’re building the kind of open-source intelligence that can rival and at times outperform centralized models on task-specific functions. This subnet isn’t just part of the network; it’s the crucible where innovation meets performance, shaping the trajectory of decentralized AI.
Leading Bittensor’s Race for Intelligence
Bittensor’s long-term viability depends on continuous breakthroughs that rival proprietary AI systems. Apex plays a central role in this mission. It’s the protocol’s flagship arena — the environment where validators and miners go head-to-head to build more capable agents, and where mechanism design is constantly being refined.
Since its rebuild in early 2024, Apex has evolved to incorporate advanced inference strategies like response ensembling and agentic workflows, adapting quickly to the demands of fast-moving LLM research. They’ve positioned themselves as the backbone for testing and scaling the very systems that will define decentralized AI’s future.
Beyond Centralized Benchmarks
What makes Apex so compelling is that they aren’t just chasing benchmarks; they’re surpassing them. By orchestrating workflows that combine LLMs with agentic reasoning, Apex has shown that open systems can outperform closed models at targeted tasks. This isn’t theoretical; it’s already being demonstrated across real-world use cases.
How Apex Incentivizes Intelligence
1. Transforming Raw Intelligence into User Products
At its core, Apex turns high-quality outputs into digital commodities. Whether it’s through chat interfaces, APIs, or enterprise tooling, users tap into a constantly evolving hivemind — one that’s trained and refined by competition, not central authority. Right now, they’re already powering four distinct products across the network.
2. Expanding the Incentive Landscape
They’ve pioneered mechanisms that reward high-quality, human-like responses — not just the ability to game synthetic prompts. By incorporating organic queries into the validation layer, Apex pushes miners to generate answers that are genuinely useful, grounded, and context-aware.
3. Fueling the Broader Bittensor Ecosystem
Apex isn’t operating in a vacuum. It’s deeply integrated with Subnet 9 for pretraining, Subnet 13 for data structuring and sentiment analysis, and Subnet 37 for fine-tuning. Together, they form a full intelligence pipeline — one that continuously creates, tests, and evolves open-source AI.
The Flagship of Open-Source AI
As the first subnet on Bittensor, Apex continues to be the standard-bearer, not just for its legacy but for what’s next. With each iteration, they’re not just refining competitive intelligence — they’re defining what decentralized AI can become.
Will Squires – CEO and Co-Founder
Will has dedicated his career to navigating complexity, spanning from designing and constructing significant infrastructure to spearheading the establishment of an AI accelerator. With a background in engineering, he made notable contributions to transport projects such as Crossrail and HS2. Will’s expertise led to an invitation to serve on the Mayor of London’s infrastructure advisory panel and to lecture at UCL’s Centre for Advanced Spatial Analysis (CASA). He was appointed by AtkinsRéalis to develop an AI accelerator, which expanded to encompass over 60 staff members globally. At XYZ Reality, a company specializing in augmented reality headsets, Will played a pivotal role in product and software development, focusing on holographic technology. Since 2023, Will has provided advisory services for the Opentensor Foundation, contributing to the launch of Revolution.
Steffen Cruz – CTO and Co-Founder
Steffen earned his PhD in subatomic physics from the University of British Columbia, Canada, focusing on developing software to enhance the detection of extremely rare events (10^-7). His groundbreaking research contributed to the identification of novel exotic states of nuclear matter and has been published in prestigious scientific journals. As the founding engineer of SolidState AI, he pioneered innovative techniques for physics-informed machine learning (PIML). Steffen was subsequently appointed as the Chief Technology Officer of the Opentensor Foundation, where he played a pivotal role as a core developer of Subnet 1, the foundation’s flagship subnet. In this capacity, he enhanced the adoption and accessibility of Bittensor by authoring technical documentation, tutorials, and collaborating on the development of the subnet template.
Michael Bunting – CFO
Before joining Macrocosmos, Mike spent 12 years in investment banking, where he guided clients through major strategic and financial transitions across more than £1 billion in international M&A and capital raising deals. Most recently serving as a Director at Piper Sandler, he brings deep experience in advising high-growth startups on strategy, business planning, funding pathways, and corporate governance. Mike has also worked closely with multinational corporations and prominent financial investors throughout his career.
Elena Nesterova – Head of Delivery
Volodymyr Truba – Senior Machine Learning Engineer
Alma Schalèn – Head of Product Design
Felix Quinque – Machine Learning Lead
Dmytro Bobrenko – Machine Learning/AI Lead
Alan Aboudib – Machine Learning Lead
Alex Williams – People & Talent Manager
Chris Zacharia – Communications Lead
Brian McCrindle – Senior Machine Learning Engineer
Lawrence Hunt – Frontend Engineer
Nicholas Miller – Senior Software Engineer
Kalei Brady – Data Scientist
Szymon Fonau – Machine Learning Engineer
Monika Stankiewicz – Executive Assistant
Amy Chai – Junior Machine Learning Engineer
Giannis Evagorou – Senior Software Engineer
Richard Wardle – Junior Software Engineer
Kai Morris – Content & Community specialist
Lewis Sword – Junior Software Engineer
Will Squires – CEO and Co-Founder
Will has dedicated his career to navigating complexity, spanning from designing and constructing significant infrastructure to spearheading the establishment of an AI accelerator. With a background in engineering, he made notable contributions to transport projects such as Crossrail and HS2. Will’s expertise led to an invitation to serve on the Mayor of London’s infrastructure advisory panel and to lecture at UCL’s Centre for Advanced Spatial Analysis (CASA). He was appointed by AtkinsRéalis to develop an AI accelerator, which expanded to encompass over 60 staff members globally. At XYZ Reality, a company specializing in augmented reality headsets, Will played a pivotal role in product and software development, focusing on holographic technology. Since 2023, Will has provided advisory services for the Opentensor Foundation, contributing to the launch of Revolution.
Steffen Cruz – CTO and Co-Founder
Steffen earned his PhD in subatomic physics from the University of British Columbia, Canada, focusing on developing software to enhance the detection of extremely rare events (10^-7). His groundbreaking research contributed to the identification of novel exotic states of nuclear matter and has been published in prestigious scientific journals. As the founding engineer of SolidState AI, he pioneered innovative techniques for physics-informed machine learning (PIML). Steffen was subsequently appointed as the Chief Technology Officer of the Opentensor Foundation, where he played a pivotal role as a core developer of Subnet 1, the foundation’s flagship subnet. In this capacity, he enhanced the adoption and accessibility of Bittensor by authoring technical documentation, tutorials, and collaborating on the development of the subnet template.
Michael Bunting – CFO
Before joining Macrocosmos, Mike spent 12 years in investment banking, where he guided clients through major strategic and financial transitions across more than £1 billion in international M&A and capital raising deals. Most recently serving as a Director at Piper Sandler, he brings deep experience in advising high-growth startups on strategy, business planning, funding pathways, and corporate governance. Mike has also worked closely with multinational corporations and prominent financial investors throughout his career.
Elena Nesterova – Head of Delivery
Volodymyr Truba – Senior Machine Learning Engineer
Alma Schalèn – Head of Product Design
Felix Quinque – Machine Learning Lead
Dmytro Bobrenko – Machine Learning/AI Lead
Alan Aboudib – Machine Learning Lead
Alex Williams – People & Talent Manager
Chris Zacharia – Communications Lead
Brian McCrindle – Senior Machine Learning Engineer
Lawrence Hunt – Frontend Engineer
Nicholas Miller – Senior Software Engineer
Kalei Brady – Data Scientist
Szymon Fonau – Machine Learning Engineer
Monika Stankiewicz – Executive Assistant
Amy Chai – Junior Machine Learning Engineer
Giannis Evagorou – Senior Software Engineer
Richard Wardle – Junior Software Engineer
Kai Morris – Content & Community specialist
Lewis Sword – Junior Software Engineer
Huge thanks to Keith Singery (aka Bittensor Guru) for all of his fantastic work in the Bittensor community. Make sure to check out his other video/audio interviews by clicking HERE.
Steffen Cruz, previously the CTO of the Opentensor Foundation, has joined forces with his longtime friend Will Squires to establish Macrocosmos. Leading subnets 1, 9, 13, 25 and 37, this team is actively shaping the future of Bittensor and stands as one of the most influential entities within the ecosystem.
In this second video, they spend much of the episode covering Subnet 13’s rebranding to “Gravity” and the team’s prediction of a Trump victory along with how this has managed to build a team of PHDs and machine learning professionals around Bittensor.
A big thank you to Tao Stats for producing these insightful videos in the Novelty Search series. We appreciate the opportunity to dive deep into the groundbreaking work being done by Subnets within Bittensor! Check out some of their other videos HERE.
In this session, the team from Macrocosmos discuss various exciting updates and innovations surrounding their subnets. They delve into the launch and functionalities of new subnets, including the Data Universe subnet for web-scale data scraping and the Apex subnet for advanced AI models. The team highlights the integration of decentralized, real-time data scraping using custom crawlers, which surpass traditional methods in efficiency and scalability. They also explore how the Bittensor community can drive decentralized AI research, leveraging massive data sets and advanced AI training. Further, the discussion touches on the evolution of scientific research within BitTensor, with the introduction of the Mainframe subnet for protein docking and drug discovery. The session concludes with an emphasis on Macrocosmos’ commitment to pushing the boundaries of decentralized AI and computational science, and the potential collaborations that can be achieved in the growing ecosystem.
This Novelty Search session, recorded in late 2024, focuses on the ongoing work and developments of the Macrocosmos team within the Bittensor ecosystem. The team dives into their progress with multiple subnets, including subnet 13, which has become one of the largest social media datasets in the world. They also discuss their work in various sectors, such as decentralized AI, data scraping, and protein folding, which is set to revolutionize fields like healthcare and pharmaceuticals. The team highlights their ongoing efforts to create a decentralized, incentive-driven system that encourages miners to contribute and optimize their computational resources. They also introduce their latest advancements in distributed training, including novel approaches to handling latency and improving gradient averaging. With a focus on data and model quality, the discussion explores how they plan to scale and refine their subnets to provide value for both validators and token holders within the Bit Tensor network. The session ends with an exciting preview of upcoming decentralized games and the future potential of the Bittensor ecosystem.
A special thanks to Mark Jeffrey for his amazing Hash Rate series! In this series, he provides valuable insights into Bittensor Subnets and the world of decentralized AI. Be sure to check out the full series on his YouTube channel for more expert analysis and deep dives.
This session, recorded in mid-2024, features a conversation with Will Squires and Stefan Cruz from Macrocosmos, a prominent player in the Bittensor ecosystem. They discuss their involvement in the decentralized AI network Bittensor, which operates through various subnets that run AI models and create computational power through a decentralized system. Will and Stefan explain their backgrounds, including their expertise in AI, machine learning, and computational research, and how these experiences led them to explore the potential of Bittensor. The conversation covers several exciting aspects, from the challenges of training AI models to the importance of creating a decentralized and transparent AI ecosystem. They also highlight the innovative and competitive nature of Bittensor, where miners are incentivized to innovate and optimize processes, driving the network forward. The discussion touches on both technical details, like the development of the protein folding subnet, and broader concepts such as the future of decentralized AI and the evolving business model of the Bittensor network.
Another video from Mark Jeffery recorded in early 2025. Mark Jeffrey sits down with Stefan and Will from Macrocosmos to discuss the launch of DAOW (the Decentralized Autonomous Oracle Wallet) and the broader changes in the Bittensor ecosystem. The team reflects on the technical success of the DAOW launch, which exceeded expectations in terms of adoption and market dynamics. They delve into the complexities of the subnet launch, including the unexpected volumes and price fluctuations, and discuss how retail investors can approach the market strategically. The conversation also covers the development of Macrocosmos’ subnets, such as their work on protein folding, data scraping, and AI training, which are all part of their “Constellation” platform. This platform aims to unite various subnets into a seamless experience, providing powerful AI tools and decentralized computing. The team emphasizes the need for both strong technical development and effective marketing to make their products accessible and impactful in the growing decentralized AI space.
Recorded in June 2026, Mark Jeffrey speaks with Will and Stefan of Macrocosmos about major progress across IOTA (Subnet 9) and Apex (Subnet 1), beginning with Orion 100B, a 100-billion-parameter distributed training run designed to prove that very large models can be trained using smaller, heterogeneous GPUs spread around the world. The team explains that IOTA uses pipeline parallelism, compression and fault-tolerant infrastructure to break models into smaller pieces, allowing commodity hardware, intermittent GPUs and even consumer Macs to contribute, with its Train at Home client downloaded around 34,000 times and peak participation reaching roughly 2,000 users. Their broader “liquid training” thesis is to make training interruptible so it can fill unused GPU capacity around higher-value inference workloads, potentially improving utilization and reducing costs for AI companies and compute providers. They say Orion reached roughly 65% of traditional data-centre training efficiency on significantly cheaper hardware and view the next step as turning IOTA into an easy-to-use commercial training platform through an SDK that can plug into existing research workflows. The conversation also covers Apex’s first external partnership with Aurelius (Subnet 37), where miners are helping research AI alignment and mechanistic interpretability by training small adapters that steer and expose the internal behaviour of language models. Macrocosmos ultimately frames IOTA as a way to democratize access to large-scale training compute and Apex as a way to democratize access to coordinated intelligence, while continuing to test larger models, cheaper hardware, mixed GPU fleets and increasingly commercial use cases.
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,
Can decentralized AI training actually compete commercially?
After revealing Project Orion at @proofoftalk, @MacrocosmosAI has spent the summer pushing the thesis further at @IOTA_SN9
Each Orion run tackles another constraint: bigger models. Heterogeneous GPUs. Interruptible
Orion’s 16B run hit a major milestone passing 100B training tokens and it's still improving.
Our co-founder @macrocrux elaborates on the achievement:
“This is the largest LLM ever pretrained using DPP: Currently at 10^22 FLOPS using heterogeneous commodity GPUs from
The run also depended on the compute providers who supplied the hardware, and who backed the idea that training can happen outside a single centralised cluster.
@green_compute_ — a UK-based, renewable-powered fleet of heterogeneous GPUs, and the largest supplier to the run.
A distributed run is, in the end, a lot of independent parts choosing to work as one system. That's true of the compute, it was just as true of the people behind it.
While the run will continue, we’d like to thank everyone, inside Macrocosmos and beyond, who made Orion's 16B run
Not bad for one day.. Baseline officially crushed and best models have almost finished the course already.
See for yourself: https://apex.macrocosmos.ai/competitions/11
If you own GPUs, you already know the number that matters isn't how many you have. It's how many are earning right now.
Allocation isn't utilisation.
Capacity can be fully booked on paper and still sit idle in practice and every idle hour is depreciation you're paying for
@IOTA_SN9 creates demand that fits the shape of that capacity, because it coordinates training across mixed, interruptible, distributed hardware, it can put compute to work that conventional training rejects.
In essence, iota turns idle or hard-to-sell capacity into billable
For a provider, this isn't about training your own models.
It's about a new revenue line on assets you've already bought. Higher productive utilisation on the fleet you own, and a way to monetise the capacity that conventional demand leaves stranded.
More of your hardware
Most companies building on AI today are building on rented ground.
The model you depend on is one someone else owns, hosts, prices, and can change. That's a reasonable trade but the more central AI becomes to how your business actually runs, the more that dependency can start
The obvious answer is to run your own model, trained on your own data, on infrastructure you control.
The reason more companies don't is that training one has been prohibitively expensive and operationally hard; the preserve of teams with reserved clusters and specialist staff.
This is what "sovereign AI" actually means at the company level: not a slogan, but owning the model your business depends on, so it can't be restricted, repriced, or switched off underneath you.
The teams that build that capability now are the ones who won't be exposed when the
