Mesh LLM vs OpenMythos

Mesh LLM is an open-source distributed inference system that pools GPU capacity across multiple machines — dense models via pipeline parallelism, MoE models via expert sharding with zero cross-node inference traffic. Every node exposes an OpenAI-compatible API, making it transparent to any existing tool or app. The standout architectural choice is Nostr-based mesh discovery: meshes are published to Nostr relays, and other nodes can discover and join them automatically with a single flag (--mesh-llm --auto). This creates a decentralized p2p compute network for running LLMs without any central registry or coordinator. Integrations with Claude Code, Goose, and other agents are built in. The project has over 800 commits and is actively maintained. For builders who want to pool compute across a homelab, a small company's GPU fleet, or even a community of friends, Mesh LLM offers the most elegant distributed inference architecture yet seen in the open-source space.

OpenMythos is a community-driven theoretical reconstruction of Claude Mythos's suspected architecture, implementing a Recurrent-Depth Transformer (RDT) — a looped transformer that recycles layers multiple times per forward pass for deeper reasoning without massive parameter growth. The project drew 10,100 GitHub stars in its first week, reflecting intense developer curiosity about what's powering Anthropic's latest generation models. The architecture has three stages: a Prelude (initial layers), a Recurrent Block (looped up to 32 times with shared weights), and a Coda (final layers). Rather than stacking hundreds of unique layers, the recurrent block runs the same weights multiple times with learned injection parameters updating hidden states between loops — enabling implicit chain-of-thought reasoning in continuous latent space without generating intermediate tokens. The project supports Grouped Query Attention (GQA) with optional Flash Attention 2, Multi-Latent Attention (MLA), and sparse MoE with routed and shared experts. Model scales range from 1B to 1T parameters. The key claim is that RDT achieves reasoning depth comparable to fixed-depth models with far more parameters, since computational complexity scales with loop iterations rather than layer count. This would explain how Claude Mythos achieves strong reasoning performance without the extreme parameter counts of brute-force scaling — though Anthropic has neither confirmed nor denied the architecture.