AWS Machine Learning Blog5858Amazon Bedrock Managed Knowledge Base achieves fine-grained access control for enterprise AI agents by injecting user context into the AgentCore Gateway tool layer and verifying JWT, combined with the ACL-aware functionality of the knowledge base.
AWS Machine Learning Blog5555AWS introduces Agent-to-Agent (A2A) protocol, allowing multi-agent systems operating across organizational boundaries to discover each other, communicate, and exchange tasks through a standardized protocol without exposing internal details or relying on the same framework.
PyTorch BlogA team of two, with the help of an AI programming agent, completed the integration of SGLang and IBM Spyre in less than a week, building a plugin with about 1,500 lines of code. SGLang's inference throughput on Spyre is about 3.6 times higher than the reference PyTorch implementation (using torch-spyre) and performs comparably to vLLM on Granite 3.3 8B and Gemma 4 26B A4B workloads.
H3C UniServer R4930 G7 server core platform has been iteratively upgraded by New H3C, equipped with a new domestic processor, featuring up to 192 cores and 768 threads per machine, with single-thread IPC performance improved by over 15%.
O'Reilly RadarAnthropic will struggle with traditional Git infrastructure by the end of 2026 when autonomous agent activities cause daily merge volumes per repository to exceed 30,000. Claude Sonnet 4.6 replaces string diffs of 2,000+ lines with AST node transformations, reducing the cost of modifying each file to O(node_delta) and enabling subsecond dependency analysis, but standard Git still discards reasoning history such as agent execution logs and compilation feedback.
The DecoderSame story · 2 outlets6161AlphaGenome 2 predicts the effects of genetic variations on specific human cell types, discovering nearly double the related gene changes compared to its predecessor. The model is trained on data encompassing 107 million cell phenotypes across 68 tissues, achieving single-cell-level precise predictions by combining DNA sequences with specific biological states as input.
Galaxy Smart Learning has launched an AI education product, Xiao Lu Ai Xue, which adopts a dual-core architecture of the Doubao large model and a self-researched intelligent learning engine. It achieves learning situation diagnosis by analyzing students' answer records and develops functions such as learning situation diagnosis, ability assessment, learning difficulty matching, and task planning. The student-side intelligent agent recommends learning tasks and explains the basis, while the teacher-side intelligent agent summarizes student performance for personalized guidance reference. The product has served over 500 schools, covering 24 provinces, and supports teachers in adjusting teaching strategies through a human-machine co-education model.
PyTorch BlogThe EUROPA Union successfully trained a MoE model with 951.9B parameters on the JUPITER supercomputer using PyTorch and TorchTitan, increasing single GPU throughput from 90 to approximately 1,300 tokens/s with 384 NVIDIA GH200 GPUs.
Hugging Face Daily PapersThrough reinforcement learning (RL) and fine-grained rewards, researchers have developed an interface called BookNavigator to help large language models (LLMs) autonomously retrieve, learn, and apply knowledge from long reference materials to solve mathematical problems.
Hugging Face Daily PapersSteerDistill quantizes LLM agents through closed-loop teacher guidance, achieving an average score 1.67-2.18 points higher than the strongest quantization baseline on W4A16, a success rate of 1.84 percentage points on τ^2-Bench, and reducing training GPU hours by 67%.
Dahua Computers has launched the FX1400 mini-server based on the Hygon 1400 processor, equipped with a 4-core 3.0GHz 15W processor, M.2 2280 PCIe slots supporting NVMe SSDs or AI acceleration cards, and measures 208×190×45mm in 3D.
vLLM BlogFeatured6060The vLLM-Omni team has integrated MiniMax H3-Image into NVIDIA Vera Rubin. With a four-card SP4 configuration, the native 4096×4096 image-to-image short prompt benchmark median latency is 0.87 seconds, and the throughput of four copies SP1×DP4 reaches 6115 images per hour.
Runway has launched the robot model Praxis-1, based on the Runway video foundation model with 27 billion parameters. In the RoboLab-120 benchmark test, it achieved a success rate of approximately 46.2%, with a speed increase of 15.6 times and a cost reduction to 1/51. This model is licensed under Apache 2.0 and both weights and code are open-sourced.
Apex Compute has released the ANR cross-platform AI rendering technology test version, which includes AI frame generation, super-resolution, and ray reconstruction features, compatible with NVIDIA, AMD, Intel graphics cards, and Steam Deck. Currently, only the AI frame generation feature is open for testing.
SpaceX is developing the Kessel AI chip rack, using a passive PCB middle board to replace the NVLink cable box, retaining 72 GPU modules, 18 computing trays, and 9 switching trays, planning to procure chip modules directly from NVIDIA and purchase the remaining components independently to bypass NVIDIA's L6 computing board and enhance supplier selection and procurement autonomy.
Hugging Face Daily PapersTOME-Math transforms 2,118 university mathematics textbooks into approximately 380K structured proof supervision records, preserving chapter order, proposition-proof links, and dependency references, used for fine-tuning to improve the mathematical reasoning ability of language models by 16.7 percentage points, surpassing all three public instruction datasets. This corpus will be released to promote structured mathematical supervision research.
Hugging Face Daily Papers6565The One Table Benchmark tested the table generation, imputation, and prediction capabilities of 46 models on 56 datasets, finding that these tasks share conditional information but have different fidelity and target interactions. Fixed models exhibit a trade-off between utility and fidelity, while frozen base models perform outstandingly in prediction and numerical imputation, indicating that unified table models have the potential to improve classification reconstruction and joint sampling.
Baidu Tianchi's 1024 card super-node is launched, equipped with optical engine technology, inter-card interconnection bandwidth increased 3 times, training performance increased 2.5 times, and inference performance increased 3 times. Accompanied by G3.5 KV Cache GPU memory acceleration cabinet, long context inference performance increased 2 times, multi-round dialogue cache hit scenario TTFT decreased by 90%, and throughput increased 7 times. A single node can support the training of large models with 500 billion parameters, with the ability to scale to ten thousand cards, and will be launched next year.
Baidu announced that its flagship AI chip, the Kunlun Core M300, will be launched next year, focusing on comprehensive performance upgrades; by 2028, multiple chips based on the new N architecture will also enter the market, precisely matching differentiated application scenarios.
Beanpod Work expands model selection, with the first batch of Kimi K3.1 Pro, GLM 5.3, and DeepSeek V4.1 Flash three third-party models. Simultaneously, the quota consumption multiplier and limited-time offers are launched. The model selection page marks the quota consumption multiplier, allowing users to weigh effectiveness, speed, and cost. During specified hours on weekdays and holidays, using participating models for activities can enjoy a discount on quota consumption. The quota consumption multiplier for Beanpod 2.1 Lite and DeepSeek V4.1 Flash is reduced.
Tongxin Intelligence has completed tens of millions of yuan in seed and angel funding within half a year, with an average chip industry experience of over 20 years for the founding team. The company designs native chips supporting vector, tensor computation, and physical intelligence based on the CUV cognitive architecture proposed by Professor Zhu Songchun. The first product is positioned as a high-value inference chip, targeting Nvidia's Thor series on the edge and B300 series on the cloud, with FPGA prototype verification planned for early December. Funds will be used for core architecture design and R&D before chip fabrication.
Hugging Face Daily PapersEmbodied First Exam has been released, including 35 sets, 47 robots, and 1,156 tasks, to evaluate the control capabilities of coding agents under physical world constraints, with GPT-6 Astra in Codex achieving a success rate of 34% in 456 core tasks under standard settings.
AWS Machine Learning BlogJava teams can leverage Amazon Bedrock and Spring AI to build autonomous AI agents, addressing scalability, governance, security, and operational challenges. Spring AI brings AI capabilities to the JVM through the Spring framework (auto-configuration, dependency injection, composable advisors), while the Amazon Bedrock AgentCore platform supports framework and model-agnostic agent construction, connection, and optimization.
OpenRouter BlogSelf-hosted LLM is only cheaper than API when GPU load is high enough. Taking the price on October 8, 2026 as an example, a 27B model needs 6-12% load on rented RTX 3090 to balance with API, while DeepSeek V4.1 Flash never balances on four H200s.
OpenRouter BlogRunning the same tasks on OpenRouter, GPT-6.1 Sol passed 180/180 tests in 60 synthetic tasks, DeepSeek had the lowest cost passing 157/180 times, but the correlation with the public ranking is low (Spearman correlation coefficient 0.109). Need to check the number of task repetitions, the number of answers, and billing verification.
OpenRouter BlogThe OpenRouter Agent SDK, by enabling doom-loop detection, prevents infinite loops in six tested models when the third identical tool call is received; combining stepCountIs(40) and maxCost(x) can limit the number of runs and cost; the credit limit of the API key can control total expenditure, these measures effectively reduce the number of tool calls (from 314 to 64) and costs (from $0.29 to $0.05).
AWS Machine Learning BlogDevelopers use OpenAI Codex for code exploration and engineering work through Amazon Bedrock. Amazon Bedrock supports Codex inference and can be integrated with AWS Identity, IAM, and observability practices. It is recommended to start with Native AWS Access for small-scale pilots, use GPT-5.6 Sol for validation, authenticate through IAM Identity Center, and configure model permissions and observability targets.
Google launches the Gemini Decisions API, driven by the Gemini Decision1 model, allowing developers to evaluate up to 1000 questions for multimodal input, scoring preset options in a single forward pass and returning probabilities and confidence levels, with end-to-end latency of 150–250 milliseconds, and a fee of 0.10 USD per million input tokens.
Supermicro Computer announces the start of shipments for the H15 server series, equipped with the AMD 6th Generation EPYC 9006 series processors (code name Venice), providing up to a 33% core count increase, 2 times PCIe bandwidth and 1.6TB/s memory bandwidth, setting multiple world records in the SPEC CPU benchmark test.
Hugging Face Daily Papers6565MACaQ is a post-training quantization method for MXFP4 W4A4, which reduces costs while maintaining accuracy in language and diffusion models by using online activation transformations and Hessian-aware refinement to independently quantize each linear layer. MACaQ achieves only a 2.3% end-to-end overhead increase on Qwen3-32B with vLLM, has a lower perplexity than DuQuant++ and MR-GPTQ, and reaches the best perplexity of Llama-3-8B in 11 times shorter quantization time.
Simon WillisonSame story · 5 outletsGoogle has released the multimodal decision-making model Gemini Decisions API, supporting text, image, audio, video, and PDF, similar to Jev and OpenAI's decision models but more confident; the price is 10 cents per million input tokens, output is free, more expensive than Jev (4.2 cents per million tokens) and OpenAI (same price)
MarkTechPostFeaturedSame story · 5 outlets7676The DecoderFeaturedSame story · 5 outlets7878Google DeepMind launches Gemini Decisions API, targeting yes/no, multiple choice, and scale scoring for closed-form judgments. The underlying model, Gemini Decision 1, can handle up to 1000 independent questions at once, covering text, images, audio, video, and PDF.
MarkTechPostSame story · 2 outlets6060Google DeepMind releases AlphaGenome 2, a DNA model that predicts the effects of specific cell type mutations. The model reads 1 Mb DNA and target cell state, significantly improving single-cell eQTL prediction performance through conditional sequence to functional architecture, increasing recall from 23% to 43% on the TenK10K benchmark, and can adapt to unseen datasets without fine-tuning.
Apple Machine Learning ResearchWeighted State Souping proposes query dependency weighting to combine cached State state encodings, significantly improving long-context performance on tasks such as single-hop and multi-hop document question answering, pinpoint retrieval, variable tracking, and tool use. The method can be adapted into an efficient sparse variant, achieving near-constant encoding costs while maintaining (or sometimes improving) performance.
Apple Machine Learning ResearchState Space Models (SSMs) lag behind Transformers in contextual learning and precise retrieval tasks, as their gating mechanism leads them to first learn memory solutions, delaying or preventing the convergence of correct contextual learning solutions, but the gating mechanism is beneficial for improving long sequence generalization.
