Skip to content
Viewing history for 2026年10月29日 周四‹ Previous dayNext day ›AccelerationBack to now

Latest picks

Products

Launches and updates of AI products, features, apps, tools, APIs and platforms.

10/29Thu
  1. vLLM78

    vLLM announces day-0 support for MistralAI's Mistral Large 4, a trillion-parameter multimodal MoE with 1M context for advanced coding, reasoning, and cybersecurity-related tasks. FP8 and NVFP4 checkpoints, with a built-in DSpark drafter. https://twitter.com/1667249535519805451/status/2115811242556137820

    Quoted postMistral AI@MistralAI

    Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:

    Why it matters

    The checkpoint formats and built-in DSpark drafter provide concrete deployment context for developers considering serving Mistral Large 4 with vLLM.

  2. NVIDIA AI81

    NVIDIA congratulates @MistralAI on the next generation of Le Chonk, a 1T-parameter MoE multimodal model trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs. In the quoted announcement, Mistral releases its weights for agentic coding and cybersecurity. NVIDIA’s day 0 support includes NVFP4 weights on @huggingface → https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B-NVFP4 Performance-optimized Dynamo for distributed inference → https://docs.nvidia.com/dynamo/recipes/mistral-large-4

    Quoted postMistral AI@MistralAI

    Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:

    Why it matters

    The post connects the released model weights with NVFP4 availability and a Dynamo inference recipe, providing concrete deployment context for teams considering running the model themselves.

  3. Hugging Face Blog71

    Introducing AnySpeechRecognizer and AnyDecisionModel for Speech and Decisions on Apple Platforms

    The authors announce two Swift packages for Apple platforms, AnySpeechRecognizer and AnyDecisionModel. AnySpeechRecognizer mirrors Apple's speech APIs while adding local Parakeet and Qwen3-ASR engines and speaker diarization for up to four speakers; AnyDecisionModel returns typed answers and probabilities for yes-or-no, multiple-choice, and rating-scale questions through local or hosted backends. Both support downloading open models from the Hugging Face Hub, and the reported speech evaluations cover macOS only, with no iPhone testing yet.

    Why it matters

    The explanation of probability calibration and option order bias offers a practical method for evaluating model-based decisions before using confidence values as application thresholds.

  4. The Decoder81

    Google opens Android development to third-party AI agents and unveils new Gemini APIs

    Google introduced Android Studio Rabbit 2 with support for more than 40 coding agents through the open Agent Client Protocol, including Claude, Codex, and Google Antigravity. Android CLI updates add Compose previews and memory leak analysis for agents. Google also announced an alpha Agentic API powered by Gemini Nano 4, a beta AppFunctions API for shared tool calling, and a Connected Apps Platform preview linking apps to Google Search and Gemini.

    Why it matters

    Direct IDE access for third-party coding agents and shared app functions illustrate how Google is extending agent integration across Android development workflows and on-device app interactions.

10/28Wed
  1. NVIDIA Blog74

    NVIDIA RTX Spark Laptops Gain Nemotron 3.5 Transcribe, Agent Controls and Gaming Updates

    NVIDIA introduced Nemotron 3.5 Transcribe alongside model support, agent resource controls and gaming updates for RTX Spark laptops. The open weight speech model handles transcription, speaker attribution, translation and audio questions, supports 40 languages and is available today on Hugging Face, with quantized versions coming soon. NVIDIA App now lets users allocate GPU resources between foreground applications and background agents, while Liquid AI’s LFM3 models support local execution through llama.cpp and vLLM.

    Why it matters

    Combining speech processing updates with GPU resource controls provides context for how local models and background agents share laptop hardware with everyday applications.

  2. vLLM71

    vLLM announces day-0 serving for NVIDIA Nemotron 3.5 Transcribe through vLLM + NeMo Speech. 40 languages. Recordings up to an hour. Promptable transcripts, speaker labels and audio Q&A. Open weights in BF16, FP8 and NVFP4. https://t.co/WhcqKqd1jf

    Quoted postNVIDIA AI@NVIDIAAI

    Meet NVIDIA Nemotron 3.5 Transcribe 🎙️ An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint. Watch it keep track of who https://t.co/PEH1PeURIv

    Why it matters

    Day-0 serving through vLLM and NeMo Speech gives deployment context for the release, alongside open weights available in BF16, FP8 and NVFP4 formats.

  3. AWS Machine Learning Blog76

    AWS announces SOCI-indexed images for faster SageMaker AI jobs

    AWS announced SOCI indexing for Amazon SageMaker AI training and processing jobs across all AWS Regions, enabling container images to load lazily. In AWS benchmarks, container readiness took 3–11 seconds, and training jobs finished 1–6 minutes sooner despite a first-epoch penalty of 3–16 seconds. Users can select pre-indexed Deep Learning Containers or index custom images; lazy loading requires Amazon ECR private repositories.

    Why it matters

    The benchmarks distinguish image download savings from first-epoch overhead, providing a concrete basis for assessing workloads whose startup code accesses different portions of a container image.

10/27Tue
  1. The Verge · AI76

    Apple is adding an AI chatbot to Pages, Keynote, and Numbers

    Apple is adding an AI assistant powered by Apple and OpenAI models to Pages, Keynote, and Numbers to help edit writing, create slides, and organize data. The assistant is rolling out in beta on Apple Intelligence-compatible devices running iOS 27, iPadOS 27, or macOS 27 or later; the apps also support MCP on macOS 27 or later for third-party AI agents. Updates to Apple’s Creator’s Studio will be free to existing subscribers in November.

    Why it matters

    Support for third-party agents through MCP adds an integration dimension to Apple’s productivity updates, extending their relevance beyond the built-in assistant’s document, slide, and spreadsheet features.

  2. OpenAI Developers81

    GPT-6.1 Terra is also coming to the API and Codex. OpenAI Developers says Terra can power apps and agents that interpret images, navigate software, and take action, offering fast computer use at a lower cost for high-volume workflows. https://t.co/kYjK7mn4Bq

    Quoted postOpenAI@OpenAI

    GPT-6.1 Terra: delivering fast and capable computer use at a lower price. GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it. https://t.co/8JTN1yonmV

    Why it matters

    The stated combination of fast computer use and lower cost makes high-volume workflows a concrete use case for the model’s planned API and Codex access.

10/25Sun
  1. The Decoder77

    Nvidia open-sources Flora with projected robot video labeling costs up to 125x lower than GPT-6 Astra

    Nvidia released Flora, an open robotics video annotation engine comprising three fine-tuned Qwen3.5 models that segment actions, describe them, and generate reward signals. Nvidia estimates labeling costs of $0.72 per video hour versus $88.49 for GPT-6 Astra, based on assumed GPU rental prices rather than production measurements. Benchmark results vary, with Flora stronger at action boundary detection and failed-step identification, while Astra leads on some description and rubric-based scoring tasks.

    Why it matters

    The benchmark breakdown puts Nvidia’s projected cost advantage in context, showing how action boundary detection, description accuracy, and reward scoring favor different models across the reported tests.

10/24Sat
10/23Fri
  1. Aravind Srinivas68

    Aravind Srinivas announced that Computer now integrates @LexnLin’s open-source skill for building websites, with 5000 credits per user to experiment. https://t.co/cMP34TwrgM @AskPerplexity says Taste Skill sets an art direction before building, drawing layout and typography references from 247 sites on Awwwards and One Page Love. Tasks using /taste-skill are free through October 31, up to 5,000 credits per user. https://t.co/XWIgQZavHo

    Quoted postComputer@AskPerplexity

    The Taste Skill by @LexnLin, now in Computer's skills marketplace, sets an art direction before Computer builds a website. Layout and type draw on 247 sites from Awwwards and One Page Love. Tasks that use /taste-skill are free through October 31, up to 5,000 credits per user. https://t.co/XWIgQZavHo

    Why it matters

    Setting an art direction before website generation gives users a concrete way to guide Computer’s visual choices through a reusable skill.

  2. vLLM72

    vLLM announces day-zero support for MiniMax M3.1 Flash from @MiniMax_AI. The post describes multi-step automation, real-codebase engineering, and visual reasoning, with native NVFP4 KV, NVFP4 MoE weights, and DSpark speculative decoding in the serving stack. https://t.co/bCzbcTASDN https://t.co/4lMCd6UC6e

    Quoted postMiniMax (official)@MiniMax_AI

    MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token. Agentic performance: A generational leap in agentic tasks, coding, and cybersecurity — 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and https://t.co/0sIsMoqIp7

    Why it matters

    The announcement connects day-one model availability with specific serving features, giving deployment teams concrete context for evaluating MiniMax M3.1 Flash support in vLLM.

  3. Aravind Srinivas67

    Aravind Srinivas says the Perplexity Decision Model can now run fully locally on a Mac, without any data leaving the machine. In the quoted announcement, @perplexitydevs describes Decider for Mac as an app running the 4B pplx-decider-flash-v1.2 model without an API key, returning a probability for questions about text or an image. https://t.co/RpOmOL9dLX

    Quoted postPerplexity Developers@perplexitydevs

    You can now run pplx-decider-flash-v1.2 locally on your Mac, without an API key. Decider for Mac is an app that runs the 4B decision model on device. Ask a question about text or an image and get a probability back, all on your machine. https://t.co/1BJ4Qv3Urr https://t.co/raBBZ26kbC

    Why it matters

    Running the decision model locally without an API key offers a concrete deployment option for text and image questions while keeping data on the user's Mac.