The framework coordinates parallel agents without a central controller.
Introducing Decentralized Language Models (DeLM)!
DeLM is a multi-agent framework that enables asynchronous, verified & reusable progress!
It makes agentic tasks more accurate and significantly cheaper. For example, it achieves 65.7% on SWE-bench Verified using Gemini 3-Flash, a ~10% jump over the best centralized alternatives at less than half the cost.
Great work led by @Mao_Yuzhen !
wait this is so cool LOL
in theory if we hillclimb RLMs maybe they become incentivized to launch code blocks in this way
At a high level, here is how DeLM differs from a centralized framework:
Introducing Decentralized Language Models (DeLM)!
DeLM is a multi-agent framework that enables asynchronous, verified & reusable progress!
It makes agentic tasks more accurate and significantly cheaper. For example, it achieves 65.7% on SWE-bench Verified using Gemini 3-Flash, a ~10% jump over the best centralized alternatives at less than half the cost.
Great work led by @Mao_Yuzhen !
@a1zhang it’s certainly one RLM strategy 👀
wait this is so cool LOL
in theory if we hillclimb RLMs maybe they become incentivized to launch code blocks in this way
DeLM in action!
At a high level, here is how DeLM differs from a centralized framework:
You can read more about DeLM and try it yourself:
Paper: https://arxiv.org/abs/2606.10662 Website: https://yuzhenmao.github.io/DeLM/
DeLM in action!
@a1zhang Have you tried the muse spark contemplating mode? I can’t tell what’s going on in the backend from what they show but it looks very interesting
@lateinteraction @a1zhang Swarlm!
@a1zhang @lateinteraction @predotdev
@lateinteraction @a1zhang What are some others? How would you rank them?
The framework coordinates parallel agents without a central controller.
Introducing Decentralized Language Models (DeLM)!
DeLM is a multi-agent framework that enables asynchronous, verified & reusable progress!
It makes agentic tasks more accurate and significantly cheaper. For example, it achieves 65.7% on SWE-bench Verified using Gemini 3-Flash, a ~10% jump over the best centralized alternatives at less than half the cost.
Great work led by @Mao_Yuzhen !