“@BarrySilbert is going to be right, $TAO is going to go up” @RaoulGMI 😏

The Bittensor $TAO ecosystem weekly update Vol.14 A 🧵

Just ship it

A single government letter switched off Fable 5 for hundreds of millions of users on Friday. Now try sending that letter to Bittensor. No CEO to call. No HQ to raid. No switch to flip. That's permissionless AI. You can't deplatform a network. for @bittensor we stand

Papers are coming. We’re expanding the team to increase our capacity across technical paper production, benchmark documentation, model analysis, and reproducibility. We want this work to be rigorous and useful to the open-source drone AI community.

Today we’re announcing a new open-source SOTA for AI drone autonomy SOTApilot hit 95.34% success on Swarm — our public, 1,000-seed benchmark for general UAV navigation across 6 procedural environments GitHub in the first reply 👇 400 submissions 25 top-model improvements 2 months Our challenge for the next 3 months: make Swarm the benchmark the world uses to measure AI drone autonomy SOTApilot begins now 🐝

Anthropic is today’s headline. Sovereignty is the real story. Sovereignty extends beyond governments. It applies to individuals, communities, and enterprises. The moment a critical technology can be restricted, controlled, or withdrawn by someone else, you’re no longer in control of your own future. This is why decentralized AI matters. bittensor:native

> be jacob steeves - @const_reborn > math cs kid from simon fraser > software engineer at google > realizes AI is owned by a handful of companies > hates it > quits to build the opposite with $TAO > takes bitcoin's idea: pay people to do useful work > points it at intelligence itself > co-founds bittensor with ala shaabana > miners produce AI > validators grade it > $TAO pays whoever's best > no lab, no gatekeeper > just an open market for intelligence > everyone calls it a science project > "why would i use this over claude" > then one day the centralized one goes dark > "claude fable is no longer available" > no warning > no vote > no appeal > millions realize they were renting intelligence > and the landlord changed the locks > suddenly a model nobody can switch off doesn't sound crazy > suddenly bittensor makes sense jacob steeves built the answer years before anyone asked the question that's the day they finally get $TAO

on sept 15 2008, lehman brothers collapsed and gave the whole world a reason to believe in $BTC on june 12 2026, a single government order forced anthropic to pull its most capable models offline for every foreign national on earth, and gave the whole world a reason to believe in bittensor:native humans only react to negative things. sad but true. in 2008 we learned the banks were the problem. in 2026 we’ll learn that one kill switch sitting inside a centralised AI lab is the problem.

The All-In crew and @jason touched on something important here. The amount of genomic data being generated is exploding. The challenge is that privacy limits how much of this data can be used to advance AI, while models are becoming increasingly cautious in how they engage with genomic questions. The future of health will increasingly depend on our ability to understand human biology. That requires infrastructure, benchmarks, and training data that allow AI to reason about genomics accurately and at scale. We’re laying the groundwork for that future at Minos.

democratizing intelligence is more important than democratizing wealth $TAO is the unstoppable alternative

don’t ever ask me again why we’re building open source and decentralised models

Are you tao pilled yet anon? bittensor:native

In times like these it bears repeating a few fundamental truths: (1) intelligence is a commodity that can be distilled and redeployed (see: Deepseek, etc.) (2) decentralized coordination networks can produce this commodity more efficiently and at larger scale than any centralized entity (see: Bitcoin) (3) open source, private, permissionless intelligence is going to be an absolutely essental, fundamental human right (see: below)

Coined the "Industrialists of Bittensor" by Bittensor Founder Const We welcomed @YumaGroup COO Greg Schvey to discuss the institutionalization of Bittensor. The story of Bittensor is the story of bottom-up capitalism, the individual developer building technologies that they think should exist, then open-sourcing them and inviting those who share the same passion, energy, and desire to co-build and commercialize. Yuma has a great role to play in this evolution. Timestamps: 00:00 Introduction 03:10 Greg's Background 07:04 The Tokenization of AI 11:11 Bottom-Up Capitalism vs. Centralized Systems 14:57 Institutionalization 21:44 Decentralized AI vs. Corporate Labs 29:48 Subnet Financial Models 43:08 Driving Mass AI Adoption

Video has unique challenges which force it to operate on the edge in production. So we turned the subnet into a vision distillation machine, and the numbers are now clear.

Endure is partnering with Synth. @SynthdataCo ( $TAO SN50 ) runs a network of AI models that simulate diverse market conditions and generate synthetic asset price path data. That is exactly the kind of intelligence a decentralized risk network is built to consume. The integration is in development. More soon.

That is what we are building, and the miners on our subnet are the engine. The incentive mechanism turned an open, adversarial competition into a model that beats the giants at its job. Point the same mechanism at the next task and it does it again.

The miners did this. We benchmarked Score 'Skills' against foundation models and frontier chat VLMs under the fairest protocol we could design. Result → alone on the accuracy-latency Pareto frontier, near real time on a laptop-class CPU, while frontier models need a datacentre to do worse. The incentive mechanism works. The frontier moves on Bittensor $TAO.

I am really excited about @0xDutchmikel's new vision People and businesses all over need models in different shapes and sizes. People come to TPN if they need a model of a certain quality and size. Today that process is guesswork, you compress a model and only find out afterwards how much intelligence you gave up, and benchmarking it properly takes time, expertise and hardware most teams do not have. TLDR, yay I made a model cheaper but I have no idea how much intelligence I lost. With TPN, users submit a bid in and say I need Kimi 2.5 but I need it compressed down and working on this kind of hardware. Miners compete via software optimizations (quantization, pruning, distillation, model merging, often stacked together) to provide this new end model The interesting part is the competition, miners are competing through never ending optimizations to make models not only smaller, but also more performant. Smaller on its own while losing fidelity is the wrong tradeoff. You set hard constraints on how much intelligence you are willing to sacrifice and the network races to find the best model inside those limits, every epoch, with validators verifying results and the network compounding knowledge with every request There has been a big focus on the next generation of models and fine tuned models but this is a whole new optimization path on compression/quality to run these models on the edge The end game is frontier intelligence that runs everywhere, your laptop, your phone, cars, custom silicon, air gapped enterprises, with TPN as the optimization layer the agentic economy routes through. The industry spent a decade making models bigger, the next phase is making them usable anywhere, and that unlock is what makes private self sovereign AI actually real.

Quasar is entering its next chapter on Bittensor SN24. We are moving toward a 10T-token decentralized training run. The idea is simple Quasar Models needs more useful training, not just bigger parameter counts. Real model quality comes from tokens, data quality, training direction, and the ability to keep improving checkpoint by checkpoint. This run starts with a 5T-token phase to produce a stronger checkpoint, then continues into another 5T-token phase, reaching 10T total trained tokens. SN24 will set the direction: the starting checkpoint, the data, the training recipe, and the evaluation system. Miners become the extra compute layer. They help Quasar train faster, improve continuously, and move forward together. this would be the largest token-scale training runs ever attempted in decentralized AI. This is how we scale Quasar. Help us train it 👇