Meta Muse Spark vs OpenMythos

Meta unveiled Muse Spark on April 8, 2026 — the first model from Meta Superintelligence Labs (MSL), led by former Scale AI CEO Alexandr Wang. It marks a dramatic break from Meta's Llama-era open-source identity: Muse Spark is fully proprietary, with only a vague promise that "future versions may be open-sourced." The model currently powers the Meta AI app, meta.ai website, and is rolling out to WhatsApp, Instagram, Facebook, Messenger, and Ray-Ban Meta AI glasses. Muse Spark is natively multimodal — it handles text and images, launches parallel subagents for complex requests, and emphasizes real-world utility: analyzing product photos for nutritional comparisons, generating full websites from descriptions, and supporting health-related image analysis with physician oversight. A private API preview is available to select partners. No benchmark data was disclosed at launch, which raised eyebrows in the community. For users, Muse Spark is accessible for free through Meta's consumer apps. For developers, the closed API is a sharp contrast to the Llama ecosystem that helped Meta build enormous developer goodwill. The model is reportedly built on significantly more efficient architecture — "an order of magnitude less compute than older midsize Llama 4 variants" — which suggests MSL's infrastructure rebuild is paying off. Whether the quality matches the ambition awaits independent evaluation.

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.