Subnet 1 - Apex
Bittensor can build the best agents
Last updated
Bittensor can build the best agents
Last updated
Subnet 1 is Bittensor’s flagship subnet, focused on building state-of-the-art open-source intelligence through decentralized competitions and cutting-edge LLM research. Bittensor requires persistent research and innovation to compete with leading proprietary models for it to continue being a viable solution to closed-source and centralized alternatives.
Produce agentic workflows with LLMs that surpass the capabilities of proprietary models at specific tasks.
Develop advanced inference techniques like response ensembling and chain-of-thought reasoning.
Prove that decentralized systems can produce SOTA AI models.
Serve as Bittensor’s main entry point for new users.
Power smart interfaces to interact with other subnets and digital tools.
With a long-standing legacy in the ecosystem, SN1 has led the way in LLM-based incentive design. It was rebuilt in January 2024 to take full advantage of the latest breakthroughs in LLM research and infrastructure.
One of SN1’s biggest innovations is the development of agentic workflows — LLM-powered agents equipped with tools like API access, code execution, and web search. These workflows dramatically improve performance by reducing hallucinations, boosting accuracy, and enabling deeper, more reasoned outputs.
Subnet 1 enhances the Bittensor environment by:
Producing highly specialised, targeted intelligence as a digital commodity for users.
Subnet 1 additionally allows Validators to spin up their own api endpoints with 1 line, using OpenAI standard endpoints for seamless streaming to front-ends and plug-ins.
Expanding Bittensor's incentive landscape and framework, building on its legacy as the protocol's flagship subnet, and continuing to push the boundaries of mechanism design.
Subnet 1 works closely with other subnets to expand its capabilities:
subnets 37, Finetuning, and subnet 9, Pre-Training, through synthetic data generation,
subnet 13, Data Universe, through data annotation and sentiment analysis,
subnet 19, Inference, offloading tasks to an inference-specific subnet.
(SN1) runs open-source language models like LLaMA and Mistral, and you can interact with it easily using the Macrocosmos SDK
With the ApexClient, you can send prompts, run chat completions, and perform live web searches — all decentralized and miner-powered.
For more details about the subnet 1 R&D work, take a look at our Substack articles:
Related resources
Creating an to relentlessly build innovative agents designed to compete with proprietary models.
Providing chat interfaces, APIs, and other mechanisms for users and businesses to interact with the intelligent hivemind. At present, subnet 1 powers multiple user-facing products, one of which being , part of our Constellation platform.
To get started, you'll need an API key. Contact @victork_1 on or @victor_ck0 on and we'll get you set up quickly 🚀.
Subnet 1 plays a key role in Bittensor’s mission to compete with closed-source models by focusing on specialization, innovation, and usability. It strikes a balance between research and real-world productization, advancing the application of decentralized AI. Check out for a variety of metrics for the competition and the model's performance quality.
