Motif 3: Technical Report

We introduce Motif 3, a decoder-only Mixture-of-Experts language model with 314 billion total parameters and 13.2 billion activated per token. Each sparse MoE layer contains 384 routed experts, with eight selected per token. This fine-grained sparsity provides substantial expert capacity while limiting computation. Motif 3 is built around Grouped Differential Latent Attention (GDLA), which integrates grouped differential attention with the compressed key-value representation of Multi-head Latent Attention. The architecture further incorporates modified manifold-constrained hyper-connections, Expert Specific PolyNorm activations, and multi-token prediction to improve optimization stability, expert specialization, and inference efficiency. We pretrain Motif 3 on approximately 12.5 trillion tokens spanning web documents, STEM, code, mathematics, multilingual content, and domain-specialized corpora. Expert-balancing and numerical-stabilization techniques support stable training at scale, while selective MXFP8 computation and communication, memory-efficient fused kernels, and window-aware context parallelism enable training with context lengths up to 256K tokens. Our post-training pipeline combines general supervised fine-tuning, six specialist teachers trained with reinforcement learning, a software-engineering teacher trained with supervised fine-tuning, and Multi-teacher On-Policy Distillation. The resulting unified model consolidates complementary capabilities in reasoning, coding, tool use, professional work, long-context understanding, calibrated abstention, and instruction following. Across a broad evaluation suite, Motif 3 demonstrates competitive performance against leading open weight models, including strong results on long-horizon agentic tasks, mathematical reasoning, scientific knowledge, and hallucination-sensitive evaluation.

Motif-2-12.7B-Reasoning: A Practitioner's Guide to RL Training Recipes

We introduce Motif-2-12.7B-Reasoning, a 12.7B parameter language model designed to bridge the gap between open-weight systems and proprietary frontier models in complex reasoning and long-context understanding. Addressing the common challenges of model collapse and training instability in reasoning adaptation, we propose a comprehensive, reproducible training recipe spanning system, data, and algorithmic optimizations. Our approach combines memory-efficient infrastructure for 64K-token contexts using hybrid parallelism and kernel-level optimizations with a two-stage Supervised Fine-Tuning (SFT) curriculum that mitigates distribution mismatch through verified, aligned synthetic data. Furthermore, we detail a robust Reinforcement Learning Fine-Tuning (RLFT) pipeline that stabilizes training via difficulty-aware data filtering and mixed-policy trajectory reuse. Empirical results demonstrate that Motif-2-12.7B-Reasoning achieves performance comparable to models with significantly larger parameter counts across mathematics, coding, and agentic benchmarks, offering the community a competitive open model and a practical blueprint for scaling reasoning capabilities under realistic compute constraints.

EvoGrader: an online formative assessment tool for automatically evaluating written evolutionary explanations

EvoGrader is a free, online, on-demand formative assessment service designed for use in undergraduate biology classrooms. EvoGrader's web portal is powered by Amazon's Elastic Cloud and run with LightSIDE Lab's open-source machine-learning tools. The EvoGrader web portal allows biology instructors to upload a response file (.csv) containing unlimited numbers of evolutionary explanations written in response to 86 different ACORNS (Assessing COntextual Reasoning about Natural Selection) instrument items. The system automatically analyzes the responses and provides detailed information about the scientific and naive concepts contained within each student's response, as well as overall student (and sample) reasoning model types. Graphs and visual models provided by EvoGrader summarize class-level responses; downloadable files of raw scores (in .csv format) are also provided for more detailed analyses. Although the computational machinery that EvoGrader employs is complex, using the system is easy. Users only need to know how to use spreadsheets to organize student responses, upload files to the web, and use a web browser. A series of experiments using new samples of 2,200 written evolutionary explanations demonstrate that EvoGrader scores are comparable to those of trained human raters, although EvoGrader scoring takes 99% less time and is free. EvoGrader will be of interest to biology instructors teaching large classes who seek to emphasize scientific practices such as generating scientific explanations, and to teach crosscutting ideas such as evolution and natural selection. The software architecture of EvoGrader is described as it may serve as a template for developing machine-learning portals for other core concepts within biology and across other disciplines.