NAll news

All news

Thursday124 items

  1. PyTorch BlogFirst-partyProduct release
    AI score60

    PyTorch integrates FlyDSL backend on AMD GPUs

    The official PyTorch blog introduces that users of AMD MI350 series GPUs can now use FlyDSL through existing operator APIs, covering dense and grouped GEMM, MXFP8/MXFP4 scaled GEMM, RMSNorm, TopK, and attention kernels provided via the AOTriton 0.14b series, without calling custom kernels. On MI355X, BF16 dense GEMM achieves a 1.10x geometric mean speedup over the faster baseline among ATen and Triton, MXFP8 and MXFP4 achieve 1.58x and 1.68x over ATen respectively, and grouped GEMM achieves 1.20x over Triton. In vLLM end-to-end A/B tests, whole-request speedups are BF16 1.13x, MXFP8 2.04x, and MXFP4 3.01x, depending on the model and concurrency.

    #Product updates#English text is being prepared.#Open-source ecosystem

  2. Pydantic AI official articleFirst-partyProduct release
    AI score60

    Pydantic AI integrates E2B sandbox as a native capability

    Pydantic AI integrates E2B as a native sandbox provider. After adding E2BSandbox() to the agent's capability list, all shell commands and file operations are executed in an isolated E2B microVM, and environment variables and SSH keys do not enter the sandbox. The sandbox can be paused after a run ends and resumed on the next run, with the filesystem and memory remaining unchanged, and can be reused across processes via a credential-free reference identifier. The article gives two examples: one is a two-round run of diagnose, pause, then fix; the other is killing a worker in a Temporal workflow and having a new worker continue and complete the fix in the same sandbox.

    Laís Carvalho avatar

    #Product updates#Agents#English text is being prepared.

  3. ZhidxProduct release
    AI score58

    OpenAI API adds a 12-hour cache retention option; caches for logged-in Codex and Work users are retained permanently

    The API service for OpenAI models will offer a 12-hour cache retention option, and caches for logged-in Codex and Work users will be retained permanently. On the API side, this option applies to GPT-5.6 and later models, with a default retention of 30 minutes; when 12 hours is selected, the cache write price rises from 1.25 times the input token price to 2.5 times, while the cache read price remains unchanged. OpenAI's documentation states that the 12 hours specifies the minimum lifetime corresponding to eligibility for cache-read billing; it does not guarantee that the stored KV tensors can be reused on every request, nor does it guarantee reduced latency.

    #Product updates#OpenAI#English text is being prepared.

  4. ZhidongxiProduct release
    AI score80

    NVIDIA releases DSX full-stack AI factory architecture and DSX POD container

    At the GTC conference in Berlin, NVIDIA released the DSX full-stack AI factory reference architecture and the DSX POD compute container; the latter pre-installs 7 Vera Rubin NVL72 liquid-cooled racks into a container, ready to plug and play upon arrival on site. Jensen Huang said that adopting this solution can compress the time from compute arriving on site to going into production from 4-5 months to about 1 month, while NVIDIA's press release said partners can shorten deployment time from 6 months to as short as 14 days. The conference also introduced the OpenShell secure agent reference architecture, the Nemotron Retriever retrieval model series, the Kumo structured data prediction model series, and, in collaboration with Agile, the Agile ONE SN humanoid robot reference design; the robot itself will be available in early 2027.

    Same story · 20 reports

    #Product updates#English text is being prepared.#English text is being prepared.

  5. ZhidongxiProduct release
    AI score67

    Tencent Yuanbao Search Opens Three Access Methods: Skill, MCP, and API

    Tencent has opened up its Yuanbao Search capabilities, offering three access methods: Skill, MCP, and API. There is a limited-time free quota of 2,000 calls per month, and in October and November, users can enjoy up to 6,000 calls per month via invitation codes. Yuanbao Search has built a hundred-billion-level index, covering 13 vertical industries such as finance, government affairs, and healthcare, and connecting to 157 authoritative sources. Its average daily call volume reaches hundreds of millions, supporting more than 50 core businesses such as WorkBuddy and ima. Tests by Zhidongxi show that a single API call takes an average of about 0.42 seconds, and after integrating DeepSeek Harness, it can complete tasks such as policy queries and paper organization.

    Same story · 4 reports

    #Product updates#RAG#Agents

  6. Unite.AIProduct release
    AI score61

    Anthropic adds continuous scanning and remediation tracking to Claude Security

    On October 21, 2026, Anthropic released the updated public beta of Claude Security, adding continuous scanning, automatic remediation tracking, per-project threat models, and a new API for Claude Enterprise customers. After the initial baseline scan, continuous scanning runs at irregular intervals based on the risk and complexity of code changes; the official documentation states that scanning is stochastic, and two scans of the same code may produce different results. Issues found are automatically marked as fixed once the remediation lands, and are automatically reopened if they reappear. The product is billed directly by token with no additional platform fee; eligible customers can receive up to $25,000 in promotional credits per month for two months, with the offer running through December 31, 2026.

    Same story · 3 reports

    #Product updates#Safety / alignment#Agents

  8. Kling AIFirst-partyProduct release
    AI score53

    Kling 4.0 launches on Magnific, supporting 30-second scenes and 10 keyframes

    Kling 4.0 is now live on Magnific, allowing users to create scenes up to 30 seconds long, guide motion with up to 10 keyframes, and pair it with native audio synthesis. The official Kling AI account reposted the news, describing it as Kling capabilities on Magnific.

    Quoted postMagnific · @magnific

    Kling 4.0 is now on Magnific Create scenes up to 30 seconds long, guide the action with up to 10 keyframes, and bring it together with native audio Your scene. Your direction https://t.co/4J68I4ZDcr

    #Product updates#Video#Image generation

  11. QbitAIProduct release
    AI score60

    Zijing Zhikang, incubated by Tsinghua AIR, releases multi-agent collaborative technology and upgrades Zijing AI Hospital

    Zijing Zhikang, incubated by Tsinghua AIR, has released multi-agent collaborative technology for the medical field and upgraded Zijing AI Hospital (Agent Hospital), offering two modes: on-demand reminders and consultation. When agents' opinions differ, the disagreements are highlighted and presented to doctors for judgment. As of July 2026, the system has built 54 AI doctors and millions of hyper-realistic AI patients, covering thousands of diseases; 18 clinical departments at Beijing Tsinghua Changgung Hospital have conducted outpatient testing, with more than 40 doctors using it regularly, improving overall diagnostic and treatment efficiency by 30%. AI doctors have achieved over 96% accuracy on the MedQA dataset and attained the highest scores in multiple international medical evaluation datasets.

    Same story · 2 reports

    #Product updates#Agents#Multimodal

  12. Mastra BlogFirst-partyProduct release
    AI score52

    Mastra Launches Mastra Observability Managed Platform

    Mastra has released the Mastra Observability managed platform, which can connect directly to Mastra framework projects, automatically trace agent runs, and view the latency, tokens, and cost of model and tool calls. The platform supports online scorers to monitor correctness, team annotation of traces to build evaluation datasets, and running evaluations in CI via Gates and Verdicts to catch regressions. It offers 100k observability events per month free with 15-day retention; beyond that, it costs $10 per 100k events.

    Alex Booker

    #Product updates#Agents#English text is being prepared.

Wednesday113 items

  1. Kling AIFirst-partyProduct release
    AI score60

    Kling 4.0 launches on the fal platform

    Kling 4.0 is now live on the fal platform. The official description calls it a next-generation native multimodal video model, with improvements in stability for complex motion, consistency of character and scene references, and precision of cinematic narrative control. It is currently available only through the fal UX and fal Agent, with API access coming soon.

    Quoted postfal · @fal

    Kling 4.0 is now live on fal. A next-generation native multimodal video model with improved stability for complex motion, stronger reference consistency across characters and scenes, and more precise control over cinematic storytelling. Available exclusively through the fal UX and fal Agent. API access coming soon.

    #Product updates#Video#English text is being prepared.

  2. Kling AIFirst-partyProduct release
    AI score56

    Kling 4.0 launches on Higgsfield

    Kling 4.0 is now live on the Higgsfield platform, supporting Omni Reference to handle characters, objects, and audio simultaneously in a single generation, and can generate 30-second scenes in one pass. This version also supports setting start, middle, and end frames to control the pacing of the visuals, with native 4K marked as coming soon.

    Quoted postHiggsfield AI 🧩 · @higgsfield

    Kling is back. Kling 4.0 is now live on Higgsfield. > Omni Reference: characters, objects and audio in one generation > 30-second scenes in a single pass > Start, middle and end frames to direct every beat > Native 4K coming soon Available now on Higgsfield. https://t.co/bRzFlhCzvs

    #Product updates#Video#English text is being prepared.

  3. Aidan GomezProduct release
    AI score50

    Cohere Releases Rerank 5, Improving Ranking Accuracy and Adding Context Focus to Control Costs

    Cohere has released Rerank 5, which it says brings significant improvements in ranking accuracy and efficiency, and is offered in Pro and Fast tiers. The new version adds features such as Context Focus, giving users better control over spending, and is also launching on Cohere Model Vault, Microsoft Foundry, and Amazon SageMaker. Aidan Gomez reposted the release announcement.

    Quoted postCohere · @cohere

    Introducing Rerank 5 A few lines of code can change how you search. With this release, we’re advancing the frontier in AI retrieval and precision. Available in Pro and Fast tiers today. https://t.co/dmFz9ykNvR

    Same story · 3 reports

    #Product updates#RAG#English text is being prepared.

  5. cnBetaProduct release
    AI score60

    NVIDIA unveils DSX POD AI factory solution, cutting on-site deployment from as long as six months to as fast as 14 days

    NVIDIA has unveiled the DSX POD AI factory deployment solution, which moves the integrated installation of computing, networking, cabling, rack-level power and liquid cooling systems to the manufacturing factory, cutting on-site deployment time from as long as six months to as fast as 14 days. The 14 days refers to deployment time on a site that is already prepared, and does not include the full cycle from site selection to completion; customers still need to complete supporting facilities such as power, cooling and networking in advance. NVIDIA plans to deliver modular AI factories based on this design worldwide in early 2027 together with partners including Dell Technologies, HPE, Lenovo, Bull, Schneider Electric and Vertiv.

    Same story · 20 reports

    #Product updates#English text is being prepared.#English text is being prepared.

  6. Cohere Product and Research BlogFirst-partyProduct release
    AI score63

    Cohere releases Rerank 5 Pro and Fast retrieval models

    Cohere releases enterprise retrieval model Rerank 5, in Pro and Fast tiers, now available on Cohere API, Model Vault, Microsoft Foundry and Amazon SageMaker, with Pro priced at $0.05 per 1M tokens and Fast at $0.02. Cohere says Rerank 5 Pro leads all tested models on four public benchmarks: ViDoRe V3, nanoMTEB, BRIGHT and BEIR, ranking first on 39 of 46 composite datasets. The new Context Focus feature can return the most relevant text snippets within a document in a single call; in official tests, removing half the context lost only 2 percentage points of answer quality.

    Abstract image

    Same story · 3 reports

    #Model releases#RAG#Agents