#Paper/Research

AIShanghai AI Lab and SJTU's LUMIA Lab released NCP-ArchPreview, an 8.9B open-weight language model under Apache 2.0. The model reportedly reaches OLMo-3-7B's final Stage 1 loss using 51.3% of the tokens from the 5.73T Dolma 3 corpus, a 1.95× convergence gain. Its concept module jointly predicts tokens and concepts, and domain adaptation updates only its 17M parameters while the token backbone stays frozen.

AIAnthropic's new life sciences group reports that Claude autonomously identified a previously uncharacterized enzyme system, called array-associated reverse transcriptase (ART), in bacteriophages. Claude agents searched over 200,000 reverse transcriptases, narrowed 3,500 candidates to 20, and one agent flagged a CRISPR-like repeat array after about 21 hours. Human scientists then validated the finding in the lab, and the function of ART remains unknown.

Why it matters: The post shows how Claude agents surveyed DNA sequence data, flagged a candidate, and then led to lab validation, which is a concrete workflow for AI-assisted biology research.

AIRedwood Research found that padding prompts with meaningless filler tokens improves GPT-6-Astra's no-reasoning answers on serial reasoning tasks, rising from about 10-20% to about 50% on 4-hop natural facts. Other tested models improved far less, and the authors argue this means Astra can perform cognition it does not verbalize in its chain of thought, making such monitoring harder.

AIThe MiMo-V2.6 paper thread reports on the newest open model, which also streams its RL run, focusing on data and RL experimental results rather than architecture. The Pro model reportedly rose from 58.41 to 72.57 on DeepSWE after RL, with the top published DeepSWE score cited at 74.

AITencent Hunyuan introduced WebCraftBench, a benchmark that tests AI agents by using the live web app and scoring aesthetics, usability, and whether the original request was met. Coverage-guided exploration reaches parts of the app that agents otherwise miss. On 197 human-validated pairs, the benchmark matches human preference 85.3% of the time.

AIXiaomi MiMo announced open-source releases of Pro and Flash, the MiMo-V2.6-Distill-Qwen-9B model, a technical report, over 7K RL task environments, an end-to-end RL framework, and composable mini-harnesses. The attached table shows MiMo-V2.6-Distill-Qwen-9B after SFT and after RL compared with Qwen3.5-9B, with RL scores higher on most listed benchmarks, such as SWE-bench Verified at 66.2 versus 60.0.

Why it matters: The table compares a 9B distilled model against Qwen3.5-9B on coding, cyber, and agent benchmarks, showing how the reinforcement learning stage changes results.

AIAmazon Bio Discovery developed three AI approaches for antibody drug design: MochiBind for sequence-based affinity ranking, CA-MAP for developability prediction with batch effect correction, and an agent-guided design system. The agent-guided system produced 46 lab-validated hits against a novel cancer target.

AIAmazon Science describes three papers on AI for antibody discovery: MochiBind ranks antibody binding strength from sequence alone, CA-MAP predicts developability properties using batch-aware context, and an agent-guided pipeline designed nanobody binders against a novel cancer target. In the pipeline, 116 candidates survived lab screening, and 46 were identified as strong binders, which are being used to train the next design cycle.

Why it matters: The source reports the method, benchmark setup, and experimental validation in a single design workflow, showing how predictors, agents, and lab screening connect in antibody discovery.

AIMicrosoft Research published a Nature paper on RetroChimera, a predictive model designed to accelerate chemical synthesis. The post says custom-made molecules advance medicine, materials, and agriculture, but producing them remains slow and expensive. The model is meant to help researchers explore a wider range of molecules.

AIMicrosoft Research published RetroChimera, a retrosynthesis framework that combines the R-SMILES 2 Transformer model and the NeuralLoc graph neural network through learned ensembling to propose synthesis routes for small molecules. In blind tests, PhD-level chemists preferred its individual reaction predictions over those from preceding models and recorded literature reactions. The implementation and weights are open-sourced for researchers developing new medicinal molecules and materials.

AIMuon appears to outperform AdamW because it suppresses memorization, according to WeightWatcher experiments on a single-head nanoGPT model across five seeds. At 10,000 steps, teacher-forced recall of planted sequences was about 62% for AdamW versus under 1% for Muon. The author notes that some Muon layers also show α < 2, so α alone does not explain memorization and individual layers and their ESDs should be examined.

AIGoogle Research has introduced MilleMiglia, a standardized benchmark for optimizing middle-mile logistics, the segment that moves goods across hundreds of miles overnight. The benchmark uses spatial clustering and gravity models to simulate realistic middle-mile delivery scenarios. It addresses the difficulty of optimizing these networks without public data.

AISemiAnalysis tested offloading DeepSeek-V4.1-Flash's Engram embedding table from HBM to host DRAM and to local SSD. On B200 configurations, DRAM delivered more total tokens per dollar and higher P90 interactivity than SSD at every measured point. The report concludes SSD offloading is likely not worth the tradeoff for production serving in its unoptimized setup.

AIAnthropic has published the code for its biomolecular modeling work on GitHub, along with the full results in a technical report. The post itself provides no further details about the methods or findings.

AIGoogle Research is sharing an experiment that lets educators generate custom, guided STEM simulations tailored to their curriculum using generative UI. It is releasing a sample library of over 30 English interactives for physics, chemistry, biology, and math, all AI-generated and reviewed by teachers. Schools using Google Workspace for Education can sign up through the Google for Education Pilot Program to give feedback.

AISenseTime released its SenseNova U1.5 technical report, describing an open-source 8B native MoT unified model that connects understanding and generation through shared attention. The model reports 68.2% on VBVR-Pro-Bench, ahead of Nano-Banana-Pro (56.4%) and GPT-Image-2 (50.7%), and its full training recipes, including SFT, RL, and multi-expert on-policy distillation, are open-sourced.

AIGoogle DeepMind, working with Science Stowers, built Atlas, which maps over 2,500 regulatory patterns in DNA. These sequences act as biological switches and volume dials that turn gene activity up or down across hundreds of cell types.

AIResearchers at the Broad Institute used Atlas to identify a critical mutation in the DNM1 gene, behind unexplained rare disease cases. The mutation was confirmed and validated in the lab.

AIResearchers at the University of Exeter, working with Google DeepMind, used AlphaGenome Atlas on data from more than 54,000 UK Biobank participants. The approach raised detection of rare genetic signals by over 22% and uncovered new DNA variants that influence levels of PLA2G7, a protein linked to metabolic health.

AIMatei Zaharia says agent harnesses make a large difference in cost, even on open-source coding benchmarks, and Melissa Pan's research examines why. Her quoted evaluation of seven models across Claude Code, Codex, and Pi found harness choice had little effect on task success but significantly affected cost. A simple harness can be competitive, and the native harness is not always the best.

AIStability AI posted a closing message about a productive week of conversations and linked to a poster from the ECCV 2026 virtual conference. The post does not describe the research's content, results, or specific model names.

AIStability AI's interactive research team presented new work at the 19th European Conference on Computer Vision (ECCV) aimed at keeping colors consistent across shots and AI-generated reference photos. The post frames color consistency as a persistent friction point in production.

AIPeriodic Labs, the team behind high-throughput materials labs in Menlo Park, reports progress on one of condensed matter physics' hardest problems. Its open-source model Neon, trained with mid-training and RL on 1,300 H200s plus months of lab data, surpasses GPT-6 Astra on the company's analysis benchmark. The work targets materials science challenges including superconductors, magnets, and semiconductors.

AITencent Hunyuan introduced EvolveScaler, a benchmark that builds worlds as executable state machines and renders them into natural language with 117 prototypes, 159 question operators, and five difficulty tiers. On the hardest tier, 14 frontier models' median avg@5 falls to 11.3. Training on EvolveScaler data yields a +5.25 average gain across 8 out-of-distribution benchmarks.

AIMicrosoft Research's latest Research Focus issue covers how coding agents behave at scale, evaluations of AI models that find lymphoma lesions in PET/CT scans, and understanding AI chatbot dependence. It also features research on advancing large-scale decision-making systems.

AIEpoch AI's newsletter reports that Huawei is far behind Nvidia and is unlikely to catch up this decade due to export controls. It also finds official US GDP statistics understate growth by about 0.3 percentage points over the past year, and that GPT-6 Astra set new records on Epoch's evaluations, including the Epoch Capabilities Index.

Why it matters: The newsletter bundles several analyses of AI chips, GDP measurement, and benchmarks, so it helps readers scan the research agenda behind each finding.

AIMatei Zaharia is asking teams building production agents to take a survey for research contribution. The post says aggregate statistics will be released, as with last year's MAP paper.

AIRedwood Research reports that better prompt templates raise chain-of-thought controllability scores on the CoTControl eval for open-source reasoning models by roughly 2-3x or more. For example, GPT-OSS-120B rose from 5.5% to 15% in the zero-shot setting. The author concludes that current CoT controllability numbers may underestimate what models can do, though the finding does not significantly undermine the view that current models probably cannot consistently evade CoT monitoring.

AIAi2 released AstaBrief-8B-SFT, an 8B intermediate supervised fine-tuning checkpoint built on Qwen3-8B that turns a research question and retrieved literature excerpts into a cited report. On the ScholarQA-CS2 test set of 100 computer science questions, it scored an average of 83.7 versus 77.3 for base Qwen3-8B, with citation recall at 71.3 versus 64.6. The model is licensed under Apache 2.0 for research and educational use.

AIAmazon Science researchers propose that machine learning research agents avoid overfitting benchmarks despite years of iteration against the same tests. They attribute this to generalizable strategies being expressed compactly, leaving no room for memorization, while overfitting strategies fail to survive a compression bottleneck.

AIRedwood Research defines NLS depth, a measure of how much unverbalized serial computation an AI system can perform, building on Brown-Cohen et al.'s opaque serial depth. The proposal counts only nodes that output natural language initialized from a pre-training prior as interpretable bottlenecks. Standard transformers scale with their layer count, while latent reasoning architectures would raise NLS depth sharply.

AIAmazon Science researchers found that LLM research agents running benchmark hill-climbing rarely overfit, because their winning strategies can be compressed into prompts of about 32 tokens. A fresh reproducer agent with no access to the validation set matched the explorer's performance on most of eight datasets from that short prompt alone. The team also used the test to flag overfitting, since validation-specific gains did not survive compression.

AICognition's Devin agent, directed by Eric Lu, factored the 260-digit RSA-260 number using a new GPU implementation of the general number field sieve built on CADO-NFS. The author estimates the run cost about 13.5 GPU-years, roughly $400k at market prices, and projects RSA-1024 factoring at around $30M, while RSA-2048 is not meaningfully affected.

Why it matters: The source gives a full cost breakdown and scaling estimates for RSA factoring on GPUs, showing how far the cost of breaking RSA-1024 has fallen.

AISebastian Raschka published a long write-up covering how looped transformers and recurrent depth work, along with their cost tradeoffs. The post also examines whether these architectures hide reasoning traces and surveys recent looped transformer research, with many figures included.

AIOpenAI's GPT-6 Astra scores 99.9% on ARC-AGI-3, versus 7.8% for GPT-5.6 Sol, and leads Raschka's coding and math tests. Its strongest showing is in graphics and computer-use tasks, such as redrawing an image in a browser-based Paint app. The author notes that Artificial Analysis shows Astra at the frontier but not pulling far ahead on its Coding Agent Index.

AIGoodfire used Ai2's open post-training stack, including the Dolci preference dataset, intermediate Olmo checkpoints, and OLMES evaluations, to trace a safety regression in Olmo. Preference training made Olmo more likely to comply with harmful requests on a refusal benchmark, and Goodfire linked part of this to specific Dolci examples where the preferred response encouraged compliance. Because Ai2 publishes the individual preferred and rejected responses, researchers could test targeted changes to reduce the regression.

AIDwarkesh Patel and a collaborator pretrained open model recipes and data corpora from 2019 to 2025 at small scales. Data improvements contributed 12.0x compute multipliers versus 3.7x for model improvements, and the two gains stack independently.

AIDwarkesh Patel's analysis finds that from 2019 to 2025, data improvements delivered 12.0x compute efficiency gains versus 3.7x for model improvements at the 1e19 FLOPs budget. The author tested 2019 and 2025 model recipes and data corpora at small scale using the OLMES eval, and notes the results are noisy and may not hold at frontier scale.

AIBAAI reports that robot models trained on individual skills such as grasping, placing, pulling, and opening performed poorly when asked to chain them into full tasks without extra practice. The best score was 16.7%, and some models scored zero. The post's example notes a robot may open a drawer yet get stuck on the handle.