Google Launches 'Sleep Framework' Continuous Learning Solution, Solving Large Model Knowledge Solidification and Forgetting Problems Through a 'Wake-Sleep' Cycle; The Model Processes Input When Awake, Integrates Memory and Self-Improves When Asleep, Experiments Show It Can Handle Millions of Token Contexts While Maintaining Stable Performance, AIME Benchmark Accuracy Improves
Meta has launched the AI fashion app Style Sketch, where users can link their Instagram accounts to generate a virtual wardrobe and receive outfit recommendations, and can also try on clothes on a virtual model. The app supports iOS and Android systems and has already launched in Indonesia, Singapore, and Thailand.
OpenRouter Blog5858Claude Sonnet 5.5 is the default choice, Claude Haiku 5.5 is suitable for high-volume and sub-agent work, Claude Opus 5.5 is used for long tasks, ambiguous or high error cost scenarios.
OpenRouter Blog5858Claude Code installation is free, the cost depends on the model called. On OpenRouter, using an API key, the median cost for 2-9 rounds of Claude Opus 5.5 is 0.31 USD, and 10-49 rounds is 1.39 USD; the corresponding cost for Claude Sonnet 5.5 is 0.16 USD and 0.60 USD.
Hugging Face Daily Papers3838Soft-MTP removes noise gradients residuals in multi-label prediction (MTP) using KL recovery techniques, improving perplexity, speculative decoding, and reasoning ability without increasing cost.
Databricks Blog4747Databricks announces the general availability of Spark Declarative Pipelines Real-Time Mode, which achieves millisecond-level end-to-end latency through continuous streaming, parallel query stages, and instant data delivery.
MarkTechPostSame story · 2 outletsBaseten released joey-1, a 2B parameter decision model, which returns labels and calibration probabilities instead of generating text, used for high-volume routing, classification, and guardrail checks, with a price of 0.01 USD per million input tokens, and weights open-sourced on Hugging Face under the MIT license
How the accounts differ:Baseten Blog states the model performs best when parameter size is ≤3B, while MarkTechPost mentions a 2B parameter model.
Cohere Blog3434Use Cohere Embed, Rerank, and Parse combined with Amazon SageMaker AI, OpenSearch Serverless, and Amazon Bedrock to search financial reports and generate answers with page-level citations.
MarkTechPost4747Prime Intellect released Prime Mega MoE, adding fast overlapping reverse transmission on the basis of DeepSeek's integrated Mega MoE core, making it available for training.
TechCrunch · AISame story · 2 outlets5555Meta quietly launched an AI fashion app called Style Sketch, which connects to Instagram or allows users to manually upload photos to build a virtual wardrobe. It recommends outfit combinations based on existing clothing and provides virtual try-on features matched to body and skin color. The app is currently only launched in Indonesia, Singapore, and Thailand, supporting iOS and Android platforms.
AWS Machine Learning Blog5555Kiro combines AI agents with codebases to migrate PHP applications to Java through a repeatable eight-stage workflow, supporting developer interactive review or AWS Transform automation, and achieving methodology unification through shared skill files.
AWS Machine Learning Blog3434Epsilon has built an intent-driven UI testing platform using Amazon Nova Act and Amazon Bedrock AgentCore Browser Tool, combining natural language test cases (such as 'confirm reward balance visibility') with existing Playwright regression test suites.
Baseten BlogSame story · 2 outlets6060Base Labs releases joey-1 decision model, performs best when parameter count ≤ 3B, single request P50 latency 20ms, cost only 0.01 USD/100M tokens, supports text and image input. This model is suitable for high decision volume, low latency-sensitive scenarios, can be used or deployed via Baseten Model APIs, weights have been open-sourced to Hugging Face.
How the accounts differ:Baseten Blog states the model performs best when parameter size is ≤3B, while MarkTechPost mentions a 2B parameter model.
Baseten Blog2222Postman's Agent Stack includes four products: Passport, Fabric, Astropods, and Fern, used for preparing APIs, controlling access, and governing agent behavior. Baseten, as Postman's recommended enterprise team reasoning partner, provides fast, reliable, and cost-effective reasoning infrastructure, supporting multiple models such as Beam 1.0 and DeepSeek V4.1 Flash, ranking top in throughput and latency.
Codex releases4545Codex 0.166.0 released, adding keyboard driver transcription selection (supporting word, line, and page operations), integrated search and configurable shortcuts, as well as consistent Markdown rendering in asynchronous questions, options, answers, and transcription responses. Also supports GPT-6.1 Terra and adds codex queue --resume feature to resume unloaded threads.
Amazon ScienceAmazon Science combines knowledge distillation (KD) and the Medusa framework to optimize the travel platform LLM. After fine-tuning, KD training, and Medusa head modification, the TinyLlama 1.1B model achieves performance close to the 70B parameter Llama-3.1 teacher in AI travel planning, intelligent filtering, and accommodation description generation scenarios, with costs reduced by two-thirds and speed increased by 10-20 times.
AWS Machine Learning Blog4242MoEngage transforms 400 internal APIs into permission-controlled intelligent agent tools through Amazon Bedrock AgentCore Gateway, solving multi-team API integration challenges. This solution eliminates the need to set up MCP servers or configure networks, reducing the time to launch new APIs from days to hours while meeting constraints such as keeping APIs within the VPC, controlling identities from the server side, and ensuring tool definitions do not occupy the reasoning context window.
AWS Machine Learning Blog4242NVIDIA has launched the 120B parameter Nemotron 3.5 Super VL model on AWS, which processes one token with 12B parameters activated, combining inference and tool use, supporting text, image, document, and video understanding.
ASUS has launched a fanless edge AI system, the MS-U320, equipped with an Intel Core 5 processor 320, providing 40 TOPS of local AI computing power. The device measures 310×202×72mm and weighs 1.2kg, featuring 1 DDR5 SO-DIMM slot and 3 M.2 slots, supporting Windows and Ubuntu 24.04.
vLLM BlogFeaturedSame story · 2 outlets7878vLLM announced support for Mistral Large 4 on the release day, which is a 1 trillion parameter, 49B token-activated MoE native multimodal model, using MLA with DSA sparse indexer, and built-in DSpark speculative decoding draft model.
Hugging Face BlogFeaturedSame story · 2 outlets8585Mistral releases multimodal MoE model Mistral Large 4 (Le Chonk), with approximately 1T total parameters, 49B activation parameters, a context window of 1M tokens, offering two official versions FP8 and NVFP4, as well as GGUF made by Unsloth and llama.cpp teams.
Why it matters: The original text provides specific scores for the 1T total parameter, 49B activation multimodal MoE model on cybersecurity and coding benchmarks, along with measured throughput for local and endpoint deployment.
Hugging Face BlogHugging Face releases Granite Switch 4.1.1, an incremental update based on 4.1, maintaining full backward compatibility. New features include multilingual speech-to-text models (supporting 10 languages with native punctuation) and a decision adapter for improved performance on the Jev decision index test set. The team achieves rapid deployment by updating only the adapter without recalculating the base model, and future support for third-party custom adapters is planned.
Microsoft has released the official version of DirectX 12 Agility SDK 1.620 and the preview version 1.722, bringing updates such as cluster geometry and PTLAS to DXR to reduce ray tracing computational overhead. On the AMD RX 9070, cluster geometry boosts performance from 1.2 FPS to 79 FPS, an increase of about 66 times, with memory usage reduced by nearly 90%. New features require support from graphics card manufacturers, currently covering NVIDIA RTX, AMD RX 9000 series, and Intel Arc graphics cards.
Hugging Face Blog5050Hugging Face releases open-source speech recognizer AnySpeechRecognizer and decision model AnyDecisionModel, supporting NVIDIA Parakeet and Alibaba Cloud Qwen3-ASR engines, and can run locally on Apple platforms.
DAYJOB: Tax evaluates the performance of AI agents in 50 professional tasks across 8 tax areas, requiring agents to complete tax filing, compliance, and advisory work. Claude Sonnet 5.5 leads the rankings with a pass rate of 10.8%, while Claude Opus 5.5 has a pass rate of 9.2%. The benchmark includes 1,789 workspace files, covering various tax scenarios and complex materials, created and reviewed by tax experts.
Tenstorrent releases cloud-based general-purpose GPU Tianhe 300 and edge product Tongyang TY1200 based on the Tianshu architecture, and compares them with NVIDIA H200, Jetson Thor T5000 on-site with random questions.
Why it matters: On-site random tests, dual-track comparison of cloud and edge against NVIDIA, and providing reproducible test numbers and software stack implementation methods.
Physical Superintelligence (PSI) claims to have achieved results on 5 open problems in quantum information theory with the help of its self-developed AI research system Emmy. Papers, Lean formal proofs, and verification code have been released on GitHub for review.
Why it matters: Readers can see a new company challenging quantum open problems with a process that separates generation from formal verification, and learn about its team and funding background.
XHEART NewCore Journey releases the X7 Ultra large model chip, achieving 700 TOPS of computing power, with mass-produced models achieving 99.99% alignment, a 80% reduction in memory bandwidth consumption, and a 100% improvement in energy efficiency. The chip incorporates a high-precision operator acceleration engine, natively supporting FP16/FP32 computing, and is the first in China to mass-produce Armv9 automotive-grade CPUs, offering a 20% performance boost and a 10% improvement in energy efficiency.
Hugging Face Daily PapersThe Prox model accelerates expert mixture inference through top-k channel selection and weighted reader, while maintaining accuracy without calibration data. Under 13% and 18% budgets, the top-k version of Prox outperformed or matched the threshold version in 40 out of 43 comparisons, including all 23 generation tasks, and avoided the repetitive loop crashes caused by calibration thresholds on Gemma 4.
Momenta CEO Cao Xudong first publicly defined the large model definition chip, which improves performance 10 times compared to general chips and can support billion-parameter models to achieve large-scale L4, driving breakthroughs in physical AI applications such as robots. XHEART New Chip Journey released a new X7 Ultra large model chip, specifically designed for large models, featuring a super-large core NPU architecture and Armv9 automotive-grade CPU.
MarkTechPostInfini-AI-Lab releases open-source framework Hermit v0.1.0, with adapters that can fine-tune Codex or Claude Code agents without altering the model or framework. Using GLM 5.3 Flash on OfficeQA, CL-Bench, and ServeLearnBench, results with adapters are better than baseline results without adapters. The MWM adapter achieves 0.795 accuracy in database exploration on CL-Bench.
Hugging Face Daily Papers6565The Sensorimotor Alignment (S²) metric finds that the alignment degree between vision and motor representations of embodied agents is positively correlated with task success rates, verifying the applicability of the Platonic representation hypothesis in the robotics field. In simulation and real robot tests, models with high S² (such as Vision-Language-Action and World Action Models) perform better and can explain the phenomenon where the VLA model ignores instructions and relies on visual features, serving as a low-cost offline proxy metric.
NewCore Voyage has released the X7 Ultra large model chip, which includes a high-precision operator acceleration engine, supports 10 times the parameter model deployment, and achieves an NPU computing utilization of 98.7%. Memory bandwidth consumption is reduced by 80%, and energy efficiency is improved by 100%. This chip is the first in China to mass produce Armv9 automotive-grade CPU, with a performance increase of 20% and an energy efficiency improvement of 10%.
Xiaomi has increased the membership credit for MiMo Desktop without increasing prices, taking effect in real-time for both new and existing members. New third-party models such as DeepSeek V4 Pro, DeepSeek V4.1 Flash, GLM 5.3, GLM 5.3 Flash, Kimi K3, Qwen 3.8 Flash, and Qwen 3.8 Max are now available for members to choose, with each model labeled with its consumption multiplier.
MolCell AI's MMDesign platform completes 18 design projects with a success rate of 94.7%, hitting all six GPCR targets among them, with the best affinity reaching nanomolar level; the platform uses a 'generate-screen-verify' workflow, generating tens of thousands of candidate molecules in computational space and testing only 9-50 to achieve success, with an overall candidate hit rate of 30.7%, demonstrating efficient and precise molecular design capabilities.
Zhuji Dynamics releases the FluxVLA Engine V2 technical report, achieving full-chain reuse of model training, evaluation, and deployment through unified configuration, modular decoupling, and system optimization. It supports 14 mainstream strategies, with inference path frequency improved by 2.31–9.64 times, validated in closed-loop execution on 6 real machines including ALOHA and dual-arm Franka. Online reinforcement learning boosts SmolVLA success rates by 10.0–15.4 percentage points.
Zhuojì Power updates the second technical report of the open-source engineering base FluxVLA Engine. V2 supports 14 strategies for LIBERO evaluation on the simulation side, increases model path frequency by 2.31–9.64 times on the inference end, and verifies deployment links on real hardware platforms such as ALOHA and dual-arm Franka. Online reinforcement learning boosts SmolVLA's success rates in LIBERO-10 and RoboTwin Clean tasks by 10.0% and 15.4%, respectively.
Diffusion Intelligence releases Clover-35B-A3B-Preview model, with 350 billion parameters achieving a median execution time of 1.52 seconds in the web Agent evaluation WindTunnel, twice as fast as the Jev+Mercury 2.5 combination, with a success rate of 98.6%. The model adopts a MoE architecture, supports thinking mode, and scores 73.6 on SWE-bench Verified.
