Distributed training has never been easy. Building a distributed system that makes economic sense on Bittensor is harder still.
Today, we’re introducing Connito: a network for collaboratively building composable specialized AI.
Signal shows us that just by training a particular selection of experts as a partial model, we can improve the performance of the whole system.
Learn about our architecture and results: https://t.co/P7IMJO7q3Z
OpenAI is Losing to Open Source.
We quietly launched on Bittensor 3 weeks ago.
Since then, 45 of our agents have beaten GPT 5.4 at one of the hardest online shopping evals made so far.
The best agent achieved a 63.0% Success Rate, performing over 15 points better than OpenAI's GPT-5.4.
We've achieved this by building the most ambitious software competition in the world.
Conventional wisdom says you need scale, billions in compute, proprietary data, and massive teams to build the best state of the art software.
But we've proven this wrong. What you actually need is competition.
Every day, armies of developers around the world study failure cases, make critical improvements in logic, and resubmit constantly. Every cooldown, agents are incrementally perfected.
Bittensor is the perfect platform for this.
Miners on the network submit agents. Validators evaluate them against real, difficult tasks. Every evaluation is public and every trajectory is inspectable. You can see exactly where every single agent succeeded and where it failed.
The top labs have poured hundreds of billions of dollars into their agents, yet ORO has been able to beat them without using any proprietary APIs. How? Every agent is open-source served through @chutes_ai.
This should seriously bother you if you work at a frontier lab.
When you have incentivized developers studying past solutions, working together, and iterating multiple times a day, you get the fastest R&D flywheel in the world.
In just 21 days, we've already proven that the collective intelligence of Bittensor, combined with a fiercely competitive environment, beats the output of any centralized funded lab.
Imagine what we'll do in a year.
Introducing Capy, the world’s first multiplayer cloud coding platform.
Capy plans, builds, tests, and reviews your code to ship self-healing PRs fully async.
PR view, review agent, computer use, Slack, Linear, ChatGPT sub, all in one tool.
Comment to get $100 of free credits.
A new era of truth and authenticity is here.
Mooncake is now Veritas.
VERITAS — Latin for truth: the pursuit of authenticity and the courage to embrace reality as it is.