How SOMA Cuts AI Agent Costs by Compressing Context :: Bittensor SN114
About SN114 SOMA
SOMA, Bittensor Subnet 114, is building a context-compression layer designed to make AI agents cheaper to run without sacrificing their ability to complete tasks. In Novelty Search 089, the Dendrite team explains how years of mining and building across Bittensor led them to SOMA, after seeing first-hand how quickly inference and token costs grow inside real AI-agent workflows. The conversation covers SOMA’s approach to compressing repeated agent context, its miner competitions, how compressed and uncompressed agent runs are compared, integrations with coding agents and model providers, and the economics behind reducing inference spend. Before diving into SOMA, the episode also explores Dendrite’s work on Teutonic and Albedo, including decentralized pre-training, model competition, and LLM-based model evaluation. Teutonic’s current work includes a 110B-parameter decentralized training run, while Albedo explores using intelligence itself as the verifier for increasingly open-ended AI tasks. 00:00 Introduction 01:51 The Dendrite story 02:20 From mining TAO to building subnets 05:18 Mining as Dendrite's training ground 08:05 Building Teutonic, Albedo and Conjecture 08:37 Teutonic: decentralized pre-training 09:45 How model competition works 14:05 The 110B decentralized training run 18:37 How useful are decentralized foundation models? 22:37 Albedo: using LLMs to evaluate AI 28:06 Albedo model competition and benchmarks 32:28 Albedo Agent API 33:08 Conjecture and open mathematical problems 36:16 Introducing SOMA 37:17 Why AI agents become so expensive 38:59 How SOMA started 40:13 Context compression explained 40:59 SOMA's first compression competition 41:26 Somarizer demo 43:05 SOMA as an AI-agent compression layer 44:29 Why coding agents are harder to compress 45:24 OpenClaw and agent-context compression 46:27 How SOMA evaluates compressors 48:13 Copilot, Codex and Claude Code 49:23 Why agent context keeps growing 50:53 How SOMA validates savings 52:18 The scale of AI-a
Read the article at Opentensor YouTube ↗
SN114 SOMA at a glance
SN114, SOMA (α token Є), is one of the independent markets on the Bittensor network, each with its own α token and its own price. Its α currently trades at 0.017651 τ (≈ $5.54), against 7,943 τ of pool liquidity. On-chain, the best-delegated validator on it yields about 64.30% APR (simple) and 90.21% APY (compounded), emissions add 0.005335 τ to the pool each block, epochs run every 360 blocks. It was registered at block #7,312,241.
More from SN114 SOMA
Taosis indexes announcements, articles, posts and releases about Bittensor subnets from their own channels and keeps a record of each. The text above is the source’s; the figures are read from the chain by Taosis.