DeepSeek V4-Pro vs OpenMythos

DeepSeek just dropped V4-Pro and V4-Flash simultaneously — and it's a statement release. V4-Pro packs 1.6 trillion total parameters in a MoE architecture with only 49B active per token, a 1-million-token context window, and a hybrid attention system (Compressed Sparse Attention + Heavily Compressed Attention) that requires just 27% of single-token inference FLOPs compared to V3.2. Both models are Apache 2.0. The hardware story is arguably the bigger news: V4 was trained entirely on Huawei Ascend 950PR chips, zero NVIDIA. That's a geopolitical and technical milestone — it validates China's domestic AI compute stack at frontier scale. The Engram Memory System gives V4 conditional context recall (94% at 128K tokens vs ~45% for V3.2), enabling genuinely long-context reasoning. V4-Flash at 284B parameters (13B active) is the cheaper, faster sibling for production use. Pricing is expected around $0.30/M tokens for Pro. The timing — released to HN today with 99+ points within hours — confirms this as an immediate conversation in the developer community about whether open-weight frontier models have finally matched proprietary ones.

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.