Jun 9, 2026 · Artificial Intelligence

How HRM-Text Achieves 1B‑Parameter, $1K Training Cost and State‑of‑the‑Art Benchmarks

HRM-Text, a 1‑billion‑parameter model trained for under two days on 16 H100 GPUs at a cost of about $1,500, uses a hierarchical recursive architecture, a focused answer‑only loss, and a PrefixLM mask to reach competitive scores on MATH, GSM8K, and ARC‑Challenge, demonstrating an efficient alternative to scaling‑only approaches.

AI benchmarkEfficient PretrainingHRM-Text

0 likes · 19 min read

How HRM-Text Achieves 1B‑Parameter, $1K Training Cost and State‑of‑the‑Art Benchmarks

PaperAgent

Apr 21, 2026 · Artificial Intelligence

OpenMythos: Rebuilding Claude Mythos with Recursive Transformers and MoE

OpenMythos is an open‑source PyTorch reimplementation of Anthropic's Claude Mythos that uses a mixed‑expert routed recurrent Transformer, introduces Recursive Depth Transformers, Multi‑Latent Attention, and several stability mechanisms, and demonstrates parameter‑efficient scaling backed by empirical studies.

AI ArchitectureClaude MythosMoE

0 likes · 6 min read

OpenMythos: Rebuilding Claude Mythos with Recursive Transformers and MoE