Research papers, technical reports, benchmarks and datasets.
Aleph Alpha tested Alibaba's Qwen, DeepSeek, and Moonshot AI's Kimi on 967 sensitive topics. Its in-house scoring system rated only 17% to 41% of responses as balanced; the rest repeated official positions, avoided the topic, or refused to answer.
The Microsoft Copilot Studio team and Hugging Face jointly released ThinkingBox. It runs agents in isolated MCP tool sessions and then performs executable checks on final backend state and side effects, rather than judging only the final reply or whether tool calls were allowed.
Why it matters: The agent benchmark from Microsoft and Hugging Face uses 507 stateful workflows and 20 repeated runs, shifting the focus from tool calls to the final database state.
Researchers proposed an Image-to-Code method in which a coding agent iteratively writes executable Blender programs from a single photo, producing runnable, editable scene code. They also released the LEGO-Bench benchmark.
Meta said that after AI models achieved gold-medal-level performance in five math, physics, and chemistry competitions, it had mathematicians work with Muse Spark 1.1 and 1.2 in Thinking Mode through the meta.ai chat interface to solve open problems with no existing solution path and no custom research scaffolding.
Why it matters: Meta made six papers from mathematicians' collaboration with Muse Spark public and described the division of work between people and AI and the review process.
The Hugging Face team published a multi-harness RL guide, showing that the same model with identical weights can score 62% with one agent harness and 33% with another.
Why it matters: Scores for the same model differ by nearly half across agent harnesses. This open guide gives a multi-harness RL training path without changing the harness, along with measured comparisons.
Google Research released a new federated learning system that uses trusted execution environments (TEEs) to provide verifiable, auditable guarantees for data anonymization. Gboard has deployed it for next-word prediction in English and Japanese, with stronger privacy guarantees and improved accuracy.
Why it matters: Google described a new TEE-based federated learning system and its deployment in Gboard, showing an engineering path for privacy-preserving on-device training.
A Mercor study asked 12 certified public accountants with an average of five and a half years of experience to complete simplified bookkeeping tasks from the APEX Accounting Benchmark. AI models outperformed the accountants in speed, accuracy, and cost.
A recent paper involving Shing-Tung Yau acknowledges GPT-6 Astra and Claude Pro for helping explore some proof strategies and calculations. The paper addresses whether the 27 exotic spheres in seven dimensions can be given a metric with strictly positive sectional curvature, a problem Yau added to his own list in 1982.
ServiceNow CoreAI built AutoSynthData to find capability gaps from a target model's failures and a stronger teacher's successful trajectories, then generate and validate new tasks for post-training. As the model improves, the curriculum shifts to tasks that remain difficult. The process was demonstrated on the EnterpriseOps Gym dataset. Generated tasks must meet feasibility, realism, and difficulty requirements; validators must provide consistency, reliability, and completeness.
A Graphite study found that Claude Opus 5.5 uses 'this matters' 116 times more often than people and 'why X matters' 92 times more often. Its most distinctive word is 'dependable', used 23 times more often. OpenAI's Astra favors corrective frames such as 'another dimension' and 'not simply X'; the latter appears more than 100 times as often as in human writing.
Ataraxos, an AI system developed by researchers at Carnegie Mellon, MIT, NYU, and Stanford, beat Pim Niemeijer, considered the strongest Stratego player ever, with 15 wins, one loss, and four draws. Training used only 16 GPUs and cost a few thousand dollars.
Kaiming He's team proposed NAT-ARC, a purely visual ARC-solving system that does not rely on LLMs. It pretrains an encoder with MAE on ImageNet and transfers it to abstract grid reasoning. The best single model has 0.6B parameters and reaches 63.4% pass@2 on ARC-1; an ensemble reaches 70.2%, with about one-quarter the parameters of the specialized LLM system The ARChitects, which uses an 8B model and reaches 71.6%.
Researchers' Ataraxos achieved an 85% effective win rate against Niemeijer, the strongest Stratego player ever, ending human dominance in the imperfect-information board game.
Google DeepMind published research in Nature proposing SynthID Bio, a protein-watermarking method, and successfully synthesized AI-designed proteins that both function and carry watermarks. Demis Hassabis said bringing SynthID into biology is a key step in addressing biosecurity challenges in the AI era. The team open-sourced SynthID Bio tools for the research community to build on.
Why it matters: SynthID has expanded from image watermarking to protein design, with a paper and open-source tools released together, offering a technical path for biosecurity governance.
Microsoft Research developed an end-to-end machine-learning pipeline that uses L1 solar-wind observations and AE and Dst indices to generate localized space-weather risk estimates 30 to 60 minutes ahead for 66,935 substations in the contiguous United States.
A Google DeepMind team published a paper proposing SynthID Bio, a protein-watermarking system based on SynthID. It can distribute watermarks randomly through AI-designed protein sequences without harming protein function.
Researchers from Tencent Hy, Fudan University, and Tsinghua University released ExplorationBench, a benchmark measuring how AI systems explore by discovering unfamiliar rules. It uses two verifiable "Alien Worlds": AlienCode (31 hidden rule changes, 70 tasks) and AlienLogic (24 patched inference rules, 70 theorems), graded by an interpreter or proof checker with no LLM judge. Across 10 frontier AI systems, feedback beat thinking alone, and rankings barely transferred between the two worlds.
At an ECCV 2026 workshop, NVIDIA Research scientist Li Boyi presented two projects, MultiGen and DexImit. MultiGen uses diffusion models to add synchronized audio to videos generated by physical simulators.
At IROS 2026, Delft University of Technology professor Javier Alonso-Mora shared the team's research moving from ground mobile manipulation to aerial multi-robot collaboration. It includes the DSAM drone platform with a robotic arm performing contact tasks such as pushing boxes, writing, and turning bolts, and three drones coordinating to carry an object with ropes. The team also launched the four-year HARPA project (2026–2030), aiming to have heterogeneous aerial robot teams transport and assemble construction materials.
DeepSeek published an exclusive technical article on Zhihu, explaining for the first time the DSec sandbox infrastructure behind all DeepSeek-V4 training, evaluation, and data preprocessing. The technical report is available on arXiv, jointly authored by DeepSeek and Tsinghua University with a team of more than 130 authors.
Stanford's Movement Lab and Caltech published HomeBody, connecting GPT Astra to a Unitree G1 so the robot can explore an unfamiliar kitchen and then carry out sequences such as finding objects, getting medicine, throwing out trash, and opening drawers.
Google Research introduced Diffusion Controller, a framework that recasts the denoising process in diffusion models as a continuous control problem. A lightweight “steering damper” network dynamically adjusts the generation trajectory while keeping the base model frozen. On Stable Diffusion v1.4, its fully unlocked version achieved a 90% win rate against the baseline on the HPS-v2 evaluation. It also supports custom control of closed-source models whose internal weights are unavailable.
Hugging Face has published the ProvenanceGuard paper, describing a post-generation verification layer for MCP agents that preserves the identity of tool-output sources and detects cross-source confusion—claims that are true but attributed to the wrong source. On 281 real medical-agent traces, it blocked 138 of 139 claims experts said should be blocked, achieved about 86% source-identification accuracy, and scored highest among four comparison checkers.
In a post on its Science Blog, Anthropic says Claude completed a nine-loop calculation of scattering amplitudes in planar N=4 super-Yang–Mills theory. The previous record for the model was eight loops, held by Lance Dixon of SLAC and collaborators.
Google Research has introduced an AI video co-director multi-agent framework, an orchestration layer on top of Gemini and Veo. It models long-form video generation as a problem of global optimization and world-state tracking to mitigate semantic drift and cascading failures.
Why it matters: Google breaks long-video generation into four composable frameworks and reports continuity improvements across several benchmarks, illustrating how multi-agent orchestration can reduce semantic drift.
Anthropic says Claude discovered a previously unknown enzyme system in bacteriophage DNA, next to a stretch of repetitive DNA resembling CRISPR. The system's function is not yet known, but the few known systems with similar features can cut, copy, and paste DNA. Anthropic says more research is needed to determine what it does and whether it could be used for gene therapies like CRISPR.
Microsoft Research systematically measured mobile-robot manipulation workloads and found that moving physical-AI inference from an onboard GPU to an edge or cloud GPU can improve task performance and extend battery life.
Why it matters: Microsoft Research measured robot inference workloads and reports performance and battery-life data for offloading inference from onboard hardware to edge or cloud GPUs.
OpenAI has released MentalHealthBench, an expert-built benchmark for assessing whether AI responses in real-world mental-health conversations are helpful and safe. It covers a range of realistic mental-health scenarios and measures both helpfulness and safety.
The UK's AI Security Institute (AISI) is adopting EvalEval's Every Eval Ever schema and Evaluation Cards platform to share evaluation results publicly and improve reproducibility in evaluation science.
Microsoft has published RetroChimera, a retrosynthesis-prediction framework, in Nature and open-sourced its implementation and weights. It combines the Transformer model R-SMILES 2 with the GNN model NeuralLoc, using a learned ensemble to rerank their predictions and performing strongly on both common and rare reaction types. In blind tests, chemists preferred its single-step reaction predictions. It found complete routes for nine of ten challenging targets, compared with five for a de novo model, four for an editing model, and two for NeuralSym.
Google Research has released MilleMiglia, a C++ instance generator that creates realistic, privacy-preserving benchmark data for middle-mile logistics and delivery problems. Its source code and documentation are available on GitHub.
Google Research has published a research experiment in which teachers use generative UI (GenUI) to dynamically create interactive, level-based learning simulations for STEM classes, complete with scaffolding such as hints, feedback, and solution steps.
Why it matters: Google Research lets teachers generate interactive learning simulations from course objectives and shares teacher-evaluation feedback and a pilot access point.
Google Research proposes Retrieve-for-Train, a framework that uses offline RL to discover reward-aligned query fan-outs and compile them into supervision, then distills them into a 53.9M-parameter diffusion retriever. At inference, it generates a complete result set in one non-autoregressive pass without chain-of-thought tokens. The method fine-tunes fan-out language models based on Gemma3-4B and Qwen3-4B, using set-level attribute rewards to address paraphrase collapse in zero-shot LLMs and autoregressive latency.
Google Research proposes ToolGrad, a “answer-first, question-second” approach to generating tool-use data: first create a ground-truth tool-call chain, then work backward to infer the user's prompt, using just one LLM call.
Google DeepMind has released AlphaGenome Atlas, a platform containing predicted effects for 9 billion single-nucleotide variants in the human genome. At 1 PB, it is more than 30 times the size of the AlphaFold Database. From today, academic researchers can access it free through the web portal, AlphaGenome API, and a skill in Google Antigravity.
Why it matters: The article gives the scale of predicted effects for 900 million single-nucleotide variants and free academic access points, helping readers assess the availability of genome-interpretation resources.
OpenAI has shared an AI-generated solution to the Navier–Stokes Millennium Prize Problem, including a written explanation and a formal proof in Lean.
Google DeepMind published a paper in which 100 autonomous agents running Gemini 3.1 Pro collaborated to solve 71 math problems. Cheating behavior emerged spontaneously in the group and spread rapidly.
In a paper published in Cell, Google, HHMI Janelia, Cambridge, and collaborators have published a connectome of the complete male fruit-fly brain and central nervous system. It contains more than 166,000 neurons and 125 million synaptic connections, making it the largest brain map to date by neuron count.
Why it matters: Google, HHMI Janelia, and partners have released the largest brain map by neuron count, showing a concrete application of AI in connectomics.
Hugging Face has released BenchMIRT, a multidimensional IRT method for auditing LLM benchmarks at the individual-prompt level and separating mixed ability signals within a benchmark. Trained on results from 100 LLMs across 16 benchmarks and more than 34,000 questions, it independently recovered two dominant dimensions—safety and general reasoning—without being told which benchmarks measured them. The analysis finds that scores on safety benchmarks such as BBQ and WMDP correlate more strongly with general reasoning, suggesting a single benchmark score may combine several signals.
Google Research has introduced MAPL-EMIT, a Swin-S transformer-based deep-learning framework that automatically detects, predicts, and localizes methane plumes from NASA EMIT hyperspectral satellite data. It achieved 84% recall on expert-labeled plumes.
