Hugging Face Daily PapersPCBRouteAgent achieves 100%, 94.9%, 100%, and 40% D2, D3-A, D3-B, D3-C cleanliness pass rates on 247 synthetic and real PCBs, being 13.9, 36, and 10 percentage points higher than the best baseline on three real-world subsets respectively.
OpenAI NewsCodeRabbit uses OpenAI models to govern and trust AI-generated code, combining Codex to increase the software delivery volume of engineering teams by 20 times.
The Verge · AISame story · 2 outlets6565Apple offers a one-time purchase option of $49.99 to extend the use of AI features in Logic Pro, which allows users to continue using Apple Intelligence features limited by reliance on OpenAI or Google Gemini cloud AI models.
Claude.dev BlogSame story · 6 outletsAnthropic has launched four Claude models for users to choose from: Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, and Claude Fable 5.2.
How the accounts differ:TechCrunch states the input price of Fable is reduced to $8 per million tokens; others do not mention the price.
Hugging Face Daily PapersConfucius4-R2T2 is a stable prefix streaming speech recognition model for real-time applications, which converts a full context ASR model into a streaming model through post-training without architecture modification. Built based on Qwen3-ASR-1.7B, R2T2 achieves near-offline recognition accuracy and low transmission delay, supports configurable streaming block sizes, and demonstrates promising cross-lingual generalization capabilities. This model and implementation have been publicly released.
Google Developers Blog5858Google Cloud releases MaxText v0.2.4, supporting long-context training on TPUs, achieving 70% MFU for sequences of 4 million tokens with the Llama 3.1 8B model on TPU 5p using MaxText Context Parallelism technology, seamlessly scalable to 14 million tokens on TPU 5p and TPU 7
AWS Machine Learning Blog5858NVIDIA Nemotron 3.5 Super VL is now available on Amazon SageMaker JumpStart, supporting text generation from image, video, and text inputs.
AWS Machine Learning BlogSame story · 6 outletsAnthropic Launches Claude Fable 5.2 and Claude Mythos 5.2 on AWS, Fable 5.2 is its most powerful general model for coding and scientific fields, supporting unsupervised operation for days; Mythos 5.2 is provided through a validation program and performs more reliably in cybersecurity and life science fields.
How the accounts differ:TechCrunch states the input price of Fable is reduced to $8 per million tokens; others do not mention the price.
The DecoderMeta releases Muse Ember, its smallest and most efficient AI model to date, designed for handling the repetitive tasks of AI agents, such as calling tools, processing small-scale coding tasks, routing requests, extracting data, and processing long documents.
Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2, significantly improving in coding, reasoning, and visual tasks, supporting various complex applications.
One Useful ThingFeatured7878Ethan Mollick, after trying out Claude Fable 5.2 before its public release, believes that according to the definition of 'achieving the level of most human experts on most available computer-completed intellectual tasks,' it is close to AGI but still has a gap.
Why it matters: The author uses six cross-domain practical cases to illustrate that the model's ability has approached general intelligence, but lacks the ability to judge whether works are good or bad.
Claude · YouTubeFeaturedSame story · 6 outlets7575Anthropic released Claude Fable 5.2, stating that it has improved in coding, reasoning, and vision, targeting the most complex and open-ended technical work. The official recommendation is to start most coding and knowledge tasks from Opus 5.5, and Fable 5.2 has been launched in the Pro, Max, Team, and Enterprise packages.
How the accounts differ:TechCrunch states the input price of Fable is reduced to $8 per million tokens; others do not mention the price.
Claude · YouTubeFeatured6868Claude Mythos 5.2 predicts the complete AAV capsid structure, which is composed of 60 identical protein chains. It writes code for an open-source folding model to handle the overlap grouping of adjacent five chains, then uses the symmetry of the capsid to assemble the 60 chains. On AAV2, its prediction deviates from the experimental structure by about the width of an atom.
Why it matters: The original text provides specific methods and accuracy of the model in real scientific research tasks, allowing readers to judge its applicable boundary in structural prediction work.
Claude · YouTubeFeaturedSame story · 6 outlets7575Anthropic releases Claude Fable 5.2, designed for the most challenging technical challenges, with improvements in coding, reasoning, and vision, suitable for the most ambitious open-ended technical work. The official recommendation is to use Opus 5.5 for most coding and knowledge tasks, and Fable 5.2 is now available in the Pro, Max, Team, and Enterprise packages.
How the accounts differ:TechCrunch states the input price of Fable is reduced to $8 per million tokens; others do not mention the price.
Why it matters: The official statement explains the positioning and applicable scenarios of Fable 5.2, and provides a choice suggestion between Opus 5.5, making it easier to decide which model to use.
NVIDIA Technical Blog6565NVIDIA Holoscan 5 achieves gravity compensation through continuous inverse dynamics calculation, allowing robotic arms to only provide torque to support their weight during operation. This technology supports robotic applications through teaching, object manipulation, or human-robot collaboration, relying on gravity compensation loops.
The Verge · AI6565Instagram releases Effect Builder feature, allowing users to create custom filters using AI prompts, supporting additional style reference images and previewing effects. Submitted filters need to pass security review before being made available to other users, and use will link to the creator's profile.
TechCrunch · AIFeaturedSame story · 6 outlets8282Anthropic releases Mythos 5.2 and Fable 5.2, claiming new models have improvements in coding, reasoning, and vision, and communicate more clearly with users. Fable's input price drops from $10 per million tokens to $8, lower than major competitor GPT-6 Astra, with a 30% faster response speed, and can support zero data retention (ZDR) policies through Enterprise Frontier Safeguards.
How the accounts differ:TechCrunch states the input price of Fable is reduced to $8 per million tokens; others do not mention the price.
AWS Machine Learning Blog3838CrossnoKaye achieves industrial physical AI on Amazon Bedrock AgentCore through its ATLAS platform and AURA agent, improving production efficiency by 11-17% and reducing energy consumption by 21%. AURA leverages AgentCore's code interpreter capabilities to run Python workloads in isolated micro VMs to verify numerical calculations, while achieving closed-loop control through edge clusters and cloud optimization systems.
Epoch AINvidia GB300 offers a 25-fold improvement in capability compared to H100 for intelligent agent inference tasks, far exceeding the 2.5-fold increase in 8-bit FLOP/s. The IPI index measures inference capability by comparing the number of requests a chip can handle simultaneously under various output speed requirements. GB300 is set as the baseline value of 1, and chips scoring 1 can serve twice as many users under the same conditions. IPI version 1 combines results from four representative models, using GB300 as the reference benchmark.
Codex releases00GitHub Copilot released version 0.169.0-alpha.6, but no specific update content was provided.
Comfy Blog6565ComfyUI v0.44 introduces multiple quality improvement features, including the new Branch Node and Stack LoRAs Node, simplifying model selection and stacking processes. Templates and models can now automatically download sample inputs for users to run directly. Additionally, the new export feature allows users to download multiple output files at once, improving work efficiency.
MarkTechPost6767Baidu Launches ERNIE Evolve Automated Research System, Automatically Selects Research Directions, Edits Code, and Runs Experiments Based on Given Codebases, Goals, and Evaluation Criteria, Retaining Only Changes Validated by Hidden Evaluators and Independent Reviewers
MarkTechPostSame story · 4 outlets6767Google releases TPU 8i Max and TPU 8tx, TPU 8i Max integrates 2 TPU 8i chips into 1 package, doubling performance; TPU 8tx uses TPU 8t chips paired with the 5th generation AMD EPYC host to achieve confidential inference, reducing latency by 2 times.
How the accounts differ:MarkTechPost states TPU 8tx uses TPU 8t chips paired with 5th gen AMD EPYC, while TechCrunch mentions it was co-developed with AMD for secret training and running AI models with direct hardware security integration.
Meta Engineering3838LLDB supports application debugging across CPU and GPU through the neutral accelerator plugin architecture of lldb-server, separating general debugging infrastructure from hardware-specific device control, presenting CPU and GPU as independent debugging processes that work together in a single debugging session. This architecture converts hardware-specific debugging APIs into concepts understood by LLDB through plugins, supporting collaborative debugging of CPUs and GPUs with different architectures, registers, memory spaces, and execution models while maintaining the neutrality of the client.
Databricks Blog4040Databricks releases Unity Catalog Dataset (UCDataset) Beta version, reducing the average training time for recommended model workloads by 10 times by streaming Unity Catalog table data directly from cloud storage to each GPU in training jobs
Baidu Wenxin first disclosed its automatic research system ERNIE Evolve, which can autonomously propose research directions, conduct experiments, and validate results. It has been used for inference optimization in Wenxin 5.5 Flash. In the low-latency scenario optimization for Kunlun Chip P800, ERNIE Evolve took about 168 hours and went through 4 rounds of direction exploration, with the optimal version's generation speed increasing to 623.6% of the baseline. The deterministic reasoning throughput optimization increased the throughput to 133.7% of the baseline in 3 days.
Why it matters: Baidu has disclosed the architecture and production environment optimization data of the automatic research system ERNIE Evolve, which can observe the actual landing points of RSI in inference optimization.
Hugging Face Daily Papers3434LowRAM compresses keys using low-rank decomposition and uses small head networks to predict key values, while retaining residual correction for the most difficult-to-reconstruct markers, achieving 4x value cache compression with near-lossless performance. Selective reconstruction can increase peak decoding throughput by 5x for 262K context under 4x KV compression. Combining quantization and eviction, LowRAM pushes compression to 26.9x while retaining 94.5% of baseline performance.
The DecoderSame story · 4 outlets5858Google released two new systems, TPU 8tx and TPU 8i Max, targeting confidential computing and high-memory demand applications, delivering a total of four TPU systems within the year. The new systems are based on existing chips, achieving rapid iteration through different packaging, networking, and memory configurations. Data in TPU 8tx remains encrypted during processing, while TPU 8i Max doubles computing power, memory, and bandwidth within a single rack, significantly improving data center performance per square meter.
How the accounts differ:MarkTechPost states TPU 8tx uses TPU 8t chips paired with 5th gen AMD EPYC, while TechCrunch mentions it was co-developed with AMD for secret training and running AI models with direct hardware security integration.
Google launches two new products, TPU 8tx and TPU 8i Max, targeting confidential computing and ultra-long context window AI applications, aiming to challenge NVIDIA's position in the AI chip market. TPU 8tx works with AMD chips, TPU 8i Max integrates two chips to provide stronger computing power. Both products are expected to launch next year.
How the accounts differ:MarkTechPost states TPU 8tx uses TPU 8t chips paired with 5th gen AMD EPYC, while TechCrunch mentions it was co-developed with AMD for secret training and running AI models with direct hardware security integration.
Google has launched two new versions of its Tensor Processing Unit (TPU) chips, competing with NVIDIA. The new chips can meet confidential computing needs and adapt to the industry's shift towards AI agents. The four TPU products are designed for different needs, and Google's chip division is accelerating its iteration speed.
Hugging Face Blog4545The HOPE method achieves better pruning effects in mixed expert language models by considering the second-order interactions between experts. It improves performance by 6.1% over the next best method on the agentic coding task at a 50% pruning rate. This method is validated on models like Qwen3.5-122B, where each MoE layer has 256 experts. After pruning, it can reduce memory usage by about 250 GB and is compatible with quantization.
Hugging Face Daily Papers3434P²-MOPD accelerates multi-teacher strategy distillation by jointly training weights and mixing coefficient α, making experts parameter priors, significantly speeding up the process.
MarkTechPost5858Reason Machines has released an open-source Git server, Aspen, designed for handling repositories with hundreds of AI agents pushing simultaneously. It independently implements the design of OpenAI Pando, with a single 8-core server capable of sustaining 567 pushes per second, and a p99 push latency of 314 milliseconds.
Hugging Face Daily Papers3131Agent evolution enhances performance through recursive self-improvement (RSI) without updating model weights, but a high baseline score does not always lead to reliable, efficient, and transferable improvements. The study evaluated the performance of four methods on general and coding tasks, finding that all sixteen evolved agents gained new task capabilities but lost some old capabilities. One general task agent improved reliability by 5.2 points but lost ten old tasks. EvoAudit achieved the best performance through analysis, mutation, and gating mechanisms.
Ji Yuan Law Release Agent Native Model NeoHorse-1.5-Flash Preview, based on Qwen3.8-Flash training, 180B total parameters, 6B activation parameters, score improvement in Agents' Last Exam, Natural Language to Repository and other evaluations. This model iterates through recursive self-improvement (RSI) and combines the OpenSquilla platform to achieve differential diagnosis of multiple models and online learning, now open for limited internal testing.
When working with Claude Code or Codex from the terminal, users often experience mental exhaustion due to multi-threaded task processing. The author manages tasks in git by organizing work in a way similar to a priority queue, with each task being a markdown file and its status as a git snapshot. The AI agent processes new tasks in a loop, runs tests, and performs browser checks, but cannot make decisions and requires the user to handle them in the 'Owner Inbox' by viewing tasks with the Claude tool on a single page, closing them after completion to process one task at a time.
Codex releases00GitHub Copilot rust-v0.169.0-alpha.4 has been released, including the latest feature improvements and fixes.
NVIDIA Newsroom3232NVIDIA has released an optimized deployment solution for South Korea's leading open-source models, Dynamo, supporting SK Telecom A.X K2, LG AI Research K-EXAONE 2.0, Upstage Solar Open 2, and Motif Technologies Motif-3, achieving up to 2x performance improvement through optimized multi-GPU coordination and communication, accelerating model production deployment.
MIT News · AI3838The FLEX platform developed by researchers from MIT and New York University enables robots to infer deep meaning from natural language instructions, enhancing their motion planning capabilities. The platform uses GPT-4o to analyze instruction patterns, allowing robots to avoid electronic devices such as laptops and phones in virtual scenarios without explicit instructions. The entire process takes about five seconds, aiming to help robots understand the underlying intent behind human instructions.
MIT News · AI5252The WorldParticle model can simulate various material behaviors, from sliding sand to folded proteins, by decomposing objects into thousands or millions of three-dimensional particles and using neural networks to predict their motion without needing to write specific equations for each material and process. The model offers higher accuracy than numerical simulation methods in simulating deformable objects, granular materials, and proteins, and can optimize computational efficiency through hybrid techniques.
