Tuesday, Nov 10

Microsoft published this on Nov 9, 2026 (source: techcommunity.microsoft.com).

Microsoft said Real-Time Multichannel Speech-to-Text is now generally available in Azure Speech, part of Microsoft Foundry, after an August public preview. The Foundry blog says the service lets contact centers and conversational AI builders transcribe stereo audio without giving up channel separation.

AWS published this on Nov 9, 2026 (source: aws.amazon.com).

AWS said Amazon Bedrock Managed Knowledge Base now supports single-shot retrieval-augmented generation and retrieval-only requests through the AgenticRetrieveStream API, alongside existing agentic retrieval. Setting maxAgentIteration to 1 runs one retrieval and one answer-generation call with no planning rounds; setting it to 0 returns matching passages with no foundation-model calls; values from 2 to 5 keep the previous agentic planning behavior. The 9 November 2026 post says all three modes can use multiple knowledge bases and walks through an illustrative internal banking-policy assistant to show when to pick each mode.

OpenAI published this on Nov 9, 2026 (source: openai.com).

OpenAI said on 9 November 2026 that ChatGPT Enterprise is now available through AWS Marketplace, so organizations can discover, buy, and manage it using their existing AWS purchasing and billing relationship and apply eligible spend toward AWS commitments. The company said AWS customers get the same models, features, and functionality available directly from OpenAI, including ChatGPT, ChatGPT Work, Codex, and dots, with new features releasing at the same time, complementing OpenAI models already on Amazon Bedrock. Eligible customers can also access ChatGPT for Healthcare and ChatGPT for Financial Services through the Marketplace.

Ollama published this on Nov 9, 2026 (source: github.com).

Ollama published v0.40.16 on 9 November 2026, adding Ripple Preview from Ricursive Intelligence, a 65B dense model built for agentic coding and runnable with ollama run ripple-preview. The release notes say Qwen3.8 Flash Next uses about 27 GiB less memory on Apple silicon, that ollama launch claude always routes Claude Code to the selected model, and that ollama launch explains when a cloud model has been retired. The /v1/responses endpoint now respects truncation instead of silently truncating by default, /v1/models returns an empty list rather than null when no models are installed, and the build adds CUDA 13.5 support for Windows on ARM plus an updated llama.cpp.

OpenAI published this on Nov 9, 2026 (source: openai.com).

OpenAI said on 9 November 2026 that it built Control Room, a central system that tracks model workloads across its research infrastructure and can stop them when safety or security concerns arise. The system connects research inference jobs, including agentic sampling for training and evaluation, and is designed to halt model output within 35 seconds of a decision to intervene. Inference engines check Control Room before fulfilling a request, and the integration fails closed so that if Control Room itself goes down, further inference is denied. OpenAI said Control Room is already deployed across research infrastructure, works with misalignment monitors, and that fail-closed coverage is live for all frontier research workloads, with remaining research jobs due to finish rolling out over the following days.

Meta published this on Nov 9, 2026 (source: research.meta.ai).

Meta Superintelligence Labs said on 9 November 2026 that it is introducing Muse Ember, its smallest and most efficient Muse model, built for the repeated work agents do all day: tool calls, coding subtasks, routing, extraction, and long-document parsing. Muse Ember is available as muse-ember-1.0 on the Meta Model API and in Muse Code at $0.10 per million input tokens, $0.004 per million cached input tokens, and $0.50 per million output tokens, with a 1M-token context window and up to 512K output tokens. Meta said it was the first model trained using Muse Wonder Full, that it delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting, and that larger models such as Muse Wonder should plan while Ember handles volume.

@runwayml on X published this on Nov 9, 2026 (source: x.com).

Runway said on 9 November 2026 that Sightline is now in the product: users can create a 360 panorama of any location, then frame shots from any angle while the environment stays consistent across those shots, and export the panorama for 3D use as a scene environment. The company pointed users to App Studio to get started. The announcement did not name an underlying model, resolution, plan availability, or API access.

@GoogleAIStudio on X published this on Nov 9, 2026 (source: x.com).

Google AI Studio said on 9 November 2026 that builders can now create apps powered by Google Maps Platform inside AI Studio. The official post says setup is one click, that describing an idea is enough for Gemini to assemble the app, and that projects can use map tiles, Places, weather, routing, 3D buildings, Street View, and more.

@GoogleWorkspace on X published this on Nov 9, 2026 (source: x.com).

Google Workspace said on 9 November 2026 that Fill with Gemini can now attach clickable, cell-level citations that point directly to web sources, so users can cross-reference, audit, and check generated data in Sheets. The official post also said capacity per request has been expanded so data-entry and research fills can cover more cells at once.

ElevenLabs published this on Nov 9, 2026 (source: elevenlabs.io).

ElevenLabs announced ElevenAgents Campaign Manager for outbound contact campaigns. Given a contact list and a goal, agents follow up over time and retain context between contacts. The announcement describes campaign management, not a guarantee of leads, appointments, or other marketing results.

@OpenAIDevs on X published this on Nov 9, 2026 (source: x.com).

OpenAI Developers announced that a dot can now connect to multiple local computers, with users choosing which computer it uses for a task. The announcement describes building and testing an app on a dedicated computer while continuing to code on a laptop; it does not announce parallel execution across computers.

@UnrealEngine on X published this on Nov 9, 2026 (source: x.com).

Epic announced the Unreal Engine 5.9 MegaJam Preview, with expanded Unreal MCP toolsets, AI coding agents inside the editor, AI-assisted content creation, and browser-based project execution. This is a preview release, and the new AI features are experimental.

Spacefast published this on Nov 9, 2026 (source: spacefast.com).

Automattic has launched Spacefast, a platform for deploying AI-coded websites and apps. Its official site confirms the product is available; TechCrunch distinguishes the formal launch from a September soft launch.

Robbyant published this on Nov 9, 2026 (source: technology.robbyant.com).

Robbyant (Shanghai Ant Robbyant Technology, also called Ant LingBot) published LingBot-Touch, a JEPA-based tactile representation model built around a shared Haptic Transformer that turns piezoresistive, magnetic, and visuotactile signals into a unified representation for dexterous manipulation. The official project page says the model was pretrained on 4,983 hours of recordings — about 1.33 billion frames across 464,000 episodes — from tactile gloves and robot hands, and that the smallest HaT-Tiny encoder has 5.46 million parameters. The company reports evaluations on 26 real-world tasks plus 12 simulated robot tasks and says the representation can transfer to unseen tactile sensors and robot hands. The tech report, code, and model weights are posted on the official page, GitHub, Hugging Face, and ModelScope. The public GitHub repository was created on 9 November 2026.

OpenAI published this on Nov 10, 2026 (source: openai.com).

OpenAI published a 10 November 2026 report drawing on new ThinkYoung research and lessons from its education partnerships in EMEA, including work with governments and school systems in Estonia, Slovakia, Greece, and Italy. The company says the collaborations share three priorities: equitable access to frontier AI tools, jointly designed AI literacy programmes, and ongoing measurement and evaluation, adapted to local languages, curricula, and institutions. Citing ThinkYoung, OpenAI reports that 79% of surveyed students said they used AI for homework or assignments, but only 47% said they had been shown how, and it argues that governments need to support teachers so they can set clearer expectations. The report organises early lessons around five themes — clear educational goals, educator preparation, local partnerships, critical AI literacy, and credible evidence — and says teacher adoption and reported time savings are encouraging early signals, while stronger evidence is still needed on whether AI improves knowledge, independent thinking, and retention.

Anthropic published this on Nov 10, 2026 (source: anthropic.com).

Anthropic published a 10 November 2026 announcement that Anthropic, PBC has launched the roadshow for its proposed initial public offering of Class A common stock. The proposed offering consists of 825,000,000 shares, of which 801,351,354 shares are being offered by Anthropic and 23,648,646 shares are being offered by certain existing stockholders; Anthropic says it will not receive any proceeds from the sale of shares by the selling stockholders. The company also expects to grant the underwriters a 30-day option to purchase up to an additional 123,750,000 Class A shares. The initial public offering price is expected to be between $70.00 and $78.00 per share, and Anthropic has applied to list its Class A common stock on the Nasdaq Global Select Market under the symbol ANT. The announcement says the proposed offering will be made only by means of a prospectus, that a registration statement has been filed with the U.S. Securities and Exchange Commission but has not yet become effective, and that the securities may not be sold, nor may offers to buy be accepted, before that registration statement becomes effective. Today's announcement, Anthropic says, does not constitute an offer to sell or a solicitation of an offer to buy the securities.

Opencode published this on Nov 10, 2026 (source: github.com).

OpenCode released v2.2.7 on 10 November 2026. The release lets Copilot accounts that have an Auto pool pick Auto in the model picker so Copilot's router can choose a model for each session at the Auto discount, falling back to the account's default chat model if routing fails. It also adds side-panel tab shortcuts, including closing tabs to the right and keyboard shortcuts to close or reopen tabs, and a Mark as unread control on tabs, Home rows, project rows, and inbox cards. The release notes a breaking change in @opencode/ai, where ImageResponse.image is removed in favour of response.images[0], and says several SDK endpoints no longer wait up to five seconds for plugins to finish loading. Other fixes include keeping remote MCP logins signed in when refresh tokens rotate, letting the read tool accept a negative offset, and charging GitHub Copilot session costs at the amount Copilot billed rather than a catalog estimate.

OpenAI published this on Nov 10, 2026 (source: openai.com).

OpenAI published a 10 November 2026 post, 'Helping caregivers use AI in everyday life,' saying OpenAI Academy is bringing free, hands-on ChatGPT workshops to caregivers in five U.S. cities to help them build practical AI skills for everyday life. The official RSS description is the wording checked here; the full article body was not independently retrieved because the OpenAI web page returned a 403 challenge, so city names and other details beyond that announcement were not added.

@elonmusk on X published this on Nov 10, 2026 (source: x.com).

In a 10 November 2026 post, Elon Musk said xAI had dumped JAX/SGLang and rewritten its entire reinforcement-learning and inference stack from scratch in C/C++, which he said took extra time. He said pre-training is done, that a lot of cleanup was needed on the supervised fine-tuning data, and that Grok 4.8 is now in reinforcement learning, with evaluation scores starting to climb. He said the company will do a soft rollout to Bot users later this week. Musk also said Grok 4.8 will bear no resemblance to 4.7, that it is not yet incredible, and that he expects it to evolve because, in his words, the foundations are built right. This is a new progress update, not a restatement of his September remarks that Grok 4.8 was a 2.5T model finishing training on a C++ stack.

@bcherny on X published this on Nov 9, 2026 (source: x.com).

On 9 November 2026, Anthropic engineer Boris Cherny said that with Fable 5.2, the longer someone has been subscribed, the more Fable resets they get, describing it as something the company wanted to do for people who have been with it the longest. In a follow-up post the same day, he said Pro users get one reset, and Max, Ultra, and Team Premium get up to three, depending on how long they have subscribed. These two posts together confirm a vendor change to Fable usage resets for those paid plans; they do not specify the size of each reset or a new reset schedule beyond that allotment.

Anthropic published this on Nov 9, 2026 (source: support.claude.com).

A Claude Help Center article modified on 9 November 2026 says Anthropic is changing how the Messages API handles thinking blocks to protect against distillation. The article says a thinking block is a record of the reasoning Claude may produce while working on a response, and that on Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5, new API accounts can no longer edit the context around a thinking block, such as the messages, tools, or system prompt, during a multi-turn conversation. On Claude Fable 5.2 and Claude Mythos 5.2, it says preserved thinking applies to all API accounts. The API now verifies that a thinking block is sent back with the same system prompt, tools, and messages that produced it, and returns an error if they do not match; developers may opt in to have thinking blocks removed from modified requests so the model responds without seeing the thinking block. Anthropic says modifying prior context has legitimate uses, which it continues to support with those adjustments, but is also a documented technique for illicit distillation prohibited by its Usage Policy and Terms of Service.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12248 on 10 November 2026. The release notes say the tag divides Vulkan GROUP_NORM workgroups into 2D to work around a maxComputeWorkGroupCount limit, in pull request #31930.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12247 on 10 November 2026. The release notes say the tag fixes out-of-bounds reads and writes on MoltenVK, including test-backend-ops sentinel mismatch errors, and sets disable_robustness for the norm_f32, l2_norm_f32, and sum_rows_f32 paths, in pull request #31890.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12246 on 10 November 2026. The release notes say that with head sizes above 128 and KV of at least 1024, scalar flash attention had used 16 rows per workgroup, and that the tag caps Vulkan scalar flash attention at 8 rows for large heads on Qualcomm, in pull request #30829.

Google Cloud published this on Nov 10, 2026 (source: cloud.google.com).

Google Cloud CEO Thomas Kurian published an announcement of Gemini for healthcare on 10 November 2026 (the Cloud blog page shows 11 November in a +09:00 timestamp, which is still 10 November in America/New_York). The post says the industry solution extends the Gemini agent introduced at Gemini at Work, and is meant to coordinate work across clinical, financial, and administrative systems organizations already run, including medical-records synthesis, pre-appointment chart preparation, and prior-authorization review, with human oversight. It describes reusable skills written by domain experts, connectors to EHR, claims, care-management, and administrative systems that preserve existing identity and access controls, and a governed platform built to support HIPAA compliance, with audit logging and verifiable citations. Google Cloud says Gemini for healthcare is available starting that day in preview.

DSPy published this on Nov 10, 2026 (source: github.com).

The stanfordnlp/dspy project released DSPy 3.5.0 on 10 November 2026, describing it as a lighter, faster foundation. The official notes say the release changes prompts, defaults, and APIs, and tell users to read the 3.5 upgrade guide and re-evaluate optimized programs. dspy.RLM, dspy.Flex, and dspy.GEPA are no longer experimental; ReActV2 now replaces ReAct as dspy.ReAct, with structured History and a typed submit tool, and the old name is deprecated until 3.6. Tools can return images, files, and other DSPy content parts on supported routes, and RLM can save and reload conversation History. LiteLLM and OpenAI’s Python library are now optional, with supported providers called through a lighter lm15 engine; the project reports that a measured default install fell from 206.7 MiB of package files in 3.3.0 to 30.9 MiB, with fewer dependencies and a faster fresh-process import. model_type="auto" is now the default, XMLAdapter is the default Adapter, disk caches are restricted by default, and RLM can crop images and send mixed text/image subqueries.

OpenAI published this on Nov 10, 2026 (source: openai.com).

OpenAI published 'Work at the Frontier: AI’s effect on freelance work' on 10 November 2026. The official RSS description says new research from OpenAI and Upwork finds shifting demand for freelance work, new skill combinations, and larger budgets and more hours for AI-related projects. The article, retrieved via an independent text preview because the OpenAI web page returned a 403 challenge, says the comparison covers 63 matched work categories, that 43 of them followed a pattern in which activities common in ChatGPT lost share on Upwork, and that the remaining 20 showed the reverse. It says that by late 2025 clients were requesting AI in more kinds of work and combining tasks that had rarely appeared together, that the mix of tasks in AI-related postings became 45% more diverse between early 2023 and late 2025, and that although posted hourly rates were similar, AI-related projects had larger fixed-price budgets and involved more hours. The paper says the findings do not point to a simple substitution story.

Google published this on Nov 10, 2026 (source: blog.google).

Google published a 10 November 2026 post saying that at Adobe MAX it announced an expanded partnership with Adobe to bring Adobe’s creative and productivity tools into everyday Google products, and Google’s models into Adobe’s tools. The post says users will soon have a direct pathway from Google Drive and Gmail into Adobe Acrobat, including an “Edit with” entry point in file previews; that Adobe is expanding its connectors in Gemini and Gemini Enterprise with Acrobat and additional creative tools, and will bring creative tools to AI Mode in Google Search; that businesses will be able to use Adobe’s CX Enterprise Coworker agent in Gemini Enterprise to design graphics, work with PDFs, and launch marketing campaigns without switching apps; and that Adobe and Google DeepMind will explore new editing capabilities, starting with an image model for Photoshop that could enable near-real-time edits.

NVIDIA published this on Nov 10, 2026 (source: blogs.nvidia.com).

NVIDIA published a 10 November 2026 blog post, coinciding with KubeCon + CloudNativeCon, saying it is contributing NVIDIA OpenShell, an open-source runtime for securing and governing autonomous agents, to the Cloud Native Computing Foundation as a CNCF Sandbox project. The post also says NVIDIA is contributing to Agent Substrate, a Google-led open-source framework for operating long-running agent workloads on Kubernetes that has been accepted into the CNCF Sandbox. NVIDIA additionally announced new components of the NVIDIA DSX AI Factory Platform: NVIDIA Kubernetes Engine, NVIDIA Kubernetes Recipes, and NVIDIA Snapshot, which it says help teams deploy, manage, and scale GPU infrastructure on Kubernetes for AI factories. This CNCF donation and platform announcement is distinct from NVIDIA’s September Open Agent Safety Platform launch that already paired OpenShell with Sentry.

QwenLM published this on Nov 10, 2026 (source: github.com).

The QwenLM/qwen-code project released Qwen Code v0.27.2 on 10 November 2026 (published_at 2026-11-10T19:22:00Z). The official notes say Web Shell can configure WeChat, QQ, and Email channels, including QR login, scan progress, identity display, and logout where supported. Mods can run one-off $.model.complete calls, rewrite displayed AssistantMessage text via ui.render, and persist plugin values with $.store. /btw side questions are replayed for context, and Managed Agent adds private follow-ups with manual or optional automatic session recaps. Pending approvals now appear in the live panel and can trigger a terminal bell or desktop notification when the terminal is unfocused. Web Shell also shows turns, output speed, total tokens, and cache hit rate under the composer, and users can import a Skill from the original Runtime workspace of a selected Session and install it explicitly without re-uploading it. The release lists a breaking dependency upgrade shipped by the SDK and Web Shell.

QwenLM published this on Nov 10, 2026 (source: github.com).

The QwenLM/qwen-code project released the TypeScript SDK tag sdk-typescript-v0.1.32 on 10 November 2026 (published_at 2026-11-10T20:18:32Z). The official notes say this SDK release bundles CLI version 0.27.2, built from the same branch or ref as the SDK, and tell users to see the commit history for changes.

Recraft published this on Nov 10, 2026 (source: recraft.ai).

Recraft published a 10 November 2026 post saying Gemini Omni Flash 1.2, Google’s newest video model, is now available in Recraft Studio. The post, by Head of Creative Production Mitya Meshcheryakov, says that compared with version 1.1 the model keeps characters and products steadier across a clip, follows camera directions with fewer unwanted cuts, and produces clearer lip-synced dialogue, including scenes with more than one speaker, and that every clip comes with native audio. In Recraft Studio, it generates clips up to 10 seconds long in 16:9 or 9:16, in the same workspace as Recraft V4.1 and the studio’s other video models. The post notes that Google released the model on 5 November 2026 under the name Gemini Omni 1.2 Flash; this item is Recraft’s integration into its studio, not a restatement of Google’s earlier general-availability announcement.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12255 on 10 November 2026. The release notes say the tag stops renaming the recurrent cache tensors in PLaMo 2, in pull request #31984.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12254 on 10 November 2026. The release notes say the tag fixes JSON Schema allOf with plain types, such as integer, generating an empty object grammar, including an empty-range error, in pull request #31834.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12253 on 10 November 2026. The release notes say that normalize_quotes_to_json had copied raw control characters verbatim while converting pythonic tool arguments to JSON, so calls with embedded line breaks produced invalid JSON that then failed when history was replayed, and that the tag escapes C0 control characters inside quoted strings, in pull request #31937.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12252 on 10 November 2026. The release notes say the tag makes the CANN backend fail early when checking whether a buffer type is supported, in pull request #31270.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12250 on 10 November 2026. The release notes say the tag adds a tiled-transpose Vulkan concat shader for a narrow case used by Qwen models, staging a 32x32 tile through shared memory when a transposed source is concatenated on dimension 0, in pull request #30591. The project reports that on a Qwen3.6-35B-A3B measurement with 30 gated-delta-net layers, the CONCAT op fell from 11877 microseconds to 957 microseconds; those figures were not independently rerun here.

Llama.cpp published this on Nov 10, 2026 (source: github.com).

The ggml-org/llama.cpp project published pre-release b12249 on 10 November 2026. The release notes say that rpc_dispatcher::start() had aborted the process when an endpoint could not be parsed, the transport failed to initialize, the socket could not connect, or the HELLO handshake failed, and that the tag makes those connect-time failures reportable instead of fatal, in pull request #30415. Mid-stream failures still assert, which the notes say is intentional.

Oura published this on Nov 10, 2026 (source: ouraring.com).

Oura published a 10 November 2026 announcement (article published_time 2026-11-10T17:36:12+00:00) saying it has entered into a definitive agreement to acquire Counsel Health, the AI-native, physician-supervised care platform it has partnered with since May 2026. The post says Counsel’s medical AI, supervised by board-certified physicians, can answer health questions, help members interpret lab results, and route them to the next step, and that Counsel physicians can diagnose, treat, order labs, and prescribe when appropriate, inside the Oura App, with member data shared only with permission. Counsel CEO Dr. Muthu Alagappan is quoted on keeping physicians in the loop for clinical decisions. Oura says Counsel will also keep working with health plans and health systems, including for people who do not wear an Oura Ring, and that the acquisition is expected to close near the end of this year. The official post does not state a purchase price.

OpenAI published this on Nov 10, 2026 (source: openai.com).

OpenAI published 'Introducing Businesses in ChatGPT' on 10 November 2026 (RSS pubDate 10:00:00 GMT). The official post, retrieved via an independent text preview because the OpenAI web page returned a 403 challenge, says Businesses in ChatGPT is launching in beta so companies can connect with the 1.2 billion people who use ChatGPT each week. Through Business Hub, businesses can claim and verify their organization, update key details, add self-service product feeds with pricing and availability, and see aggregated insights such as mentions, website clicks, and discovery topics, which OpenAI says are shown only when thresholds are met and without sharing user conversations. After claiming a business, eligible organizations can build, test, and launch a Business Agent; when a business appears in a model response, people can open its ChatGPT profile and, if an agent is available, start a clearly labeled conversation that is separate from the original chat and from ChatGPT’s own answers. Business Agents can also be sponsored as a destination for ChatGPT Ads. OpenAI says Business Agents are now available to eligible ecommerce and online service businesses in the United States, that Shopify merchants can connect a store from Business Hub or the Shopify admin, and that it plans to expand to more markets and categories over time.

Google published this on Nov 10, 2026 (source: blog.google).

Google Product Director for Android AI Karthik Thirumalai published a 10 November 2026 post (published_time 2026-11-10T19:00:00+00:00) saying Google is introducing new Gemini Intelligence features and expanding the system to more of Android’s advanced devices. The post says Motorola is introducing its first Gemini Intelligence device, motorola signature 27, available for pre-order in the U.S. on 19 November, and that with that day’s Pixel Drop, Pixel 10, Pixel 10 Pro XL, and Pixel 10 Pro Fold get the latest Gemini Intelligence features. Rambler, which turns spoken thoughts into organized text in real time, is now available on those Pixel 10 phones and on motorola signature 27, and task automation — asking Gemini to handle logistics such as ordering food, hailing rides, or booking reservations — is expanding to the UK, India, and Japan. Create My Widget, rolling out over the following weeks, lets people describe a personalized homescreen widget in their own words, pulling in information from Gmail, Calendar, and Search, and is also coming to Pixel Watch. Device Assistance lets people ask the phone to clear space, review which apps can access location, or mute distractions while still allowing important calls. The post also says a Pixel phone with Gemini Intelligence will soon bring proactive suggestions to Android Auto.

Hugging Face published this on Nov 10, 2026 (source: huggingface.co).

Hugging Face published a 10 November 2026 blog post saying organizations can now link a Hugging Face organization to a Google Cloud account so Hub usage appears on the existing Google Cloud invoice instead of a corporate credit card. The post, by Zaïd Alhadad, Pierric Cistac, Simon Pagezy, and Michelle Habonneau, says an admin subscribes to the Hugging Face Hub listing on the Google Cloud Marketplace, links an organization they administer, and, after Hugging Face reviews the subscription, the linked Google Cloud account becomes the organization’s payment method. Coverage includes upgrading to a Team or Enterprise subscription, pay-as-you-go compute such as Jobs, Inference Endpoints, Inference Providers, GPU Spaces, and ZeroGPU quota extension, and additional storage at public Hugging Face pricing. While a subscription is pending, usage continues to be billed to any existing payment method; organizations that already have a card on file switch at the start of the next billing period, and free-tier organizations with no card usually switch immediately. The post also says adding credits is not available while marketplace billing is active, and that included credits still apply.

Kiro published this on Nov 10, 2026 (source: kiro.dev).

The Kiro changelog for IDE 1.4.154, dated 10 November 2026, says the release adds Approve for me to chat: Kiro approves routine actions, blocks risky ones, and asks when it is unsure. Where the feature is available, the Autopilot toggle becomes a permission mode menu with Supervised, Accept file changes (formerly Autopilot), and Approve for me. Existing deny, ask, and allow rules still decide first; for an action no rule covers, a model approves one it judges safe, blocks one it judges unsafe, and asks the user when unsure. Users can turn on Use Approve for me by default for new chats, or run Kiro: Enable Approve For Me (User) from the Command Palette. The same entry also lists organization-managed settings, Memory view sorting and filtering, and workflow run improvements, plus fixes for stopped turns, spec tasks, and workspace trust. A 6 November changelog already described Approve for me on Kiro Web; this item is the 10 November IDE 1.4.154 release.

GitHub published this on Nov 10, 2026 (source: github.blog).

GitHub's 10 November 2026 changelog says the Microsoft C++ Language Server for GitHub Copilot CLI now bundles its MSBuild compile-commands extractor and runs it automatically, so an agent no longer has to discover and invoke that step itself.

Google-gemini published this on Nov 10, 2026 (source: github.com).

Google published Gemini CLI v0.68.0 on GitHub on 10 November 2026 (published_at 2026-11-10T21:17:07Z). The stable tag is not a prerelease.

OpenAI published this on Nov 10, 2026 (source: github.com).

OpenAI published Codex rust-v0.167.1 (release name 0.167.1) on GitHub on 10 November 2026 (published_at 2026-11-10T21:32:01Z), following rust-v0.167.0.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12262 on 10 November 2026 (published_at 2026-11-10T23:04:37Z). The notes say the CUDA backend skips a graph compatibility check when a graph is being reused.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12261 on 10 November 2026 (published_at 2026-11-10T22:50:10Z). The notes say the mtmd path shares a compute buffer among subcontexts and adds a load-time allocation for audio generation.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12259 on 10 November 2026 (published_at 2026-11-10T22:11:56Z). The notes say the mtmd path adds Kimi K3.1 vision support, including preprocessing shared with K2.5 and the projector MLP.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12257 on 10 November 2026 (published_at 2026-11-10T21:54:51Z). The notes say CUDA graph and concurrent stream state were moved out of common.cuh so changes no longer rebuild every CUDA translation unit.

@ClaudeDevs on X published this on Nov 10, 2026 (source: x.com).

The Claude Developers account said paid Pro, Max, Ultra, and Team users each get one additional full usage reset, covering all models and limits, that can be used until 2 p.m. Pacific Time on 10 December 2026, via Settings > Usage or clau.de/reset. The post says the extra reset is to offset recent issues, including cases where a prior Fable-related reset did not help users who had already exhausted weekly limits across models. This is a separate action from the already published Official Fable usage-reset item.

GitHub published this on Nov 10, 2026 (source: github.blog).

GitHub's 10 November 2026 changelog says enterprise GitHub Apps can call the enterprise-level Copilot seat-management APIs to control which users and enterprise teams are assigned Copilot seats.

@ElevenLabs on X published this on Nov 10, 2026 (source: x.com).

ElevenLabs said on 10 November 2026 (18:10 UTC) that Scribe v3 and Scribe v3 Live are its most accurate speech-to-text models yet, aimed at noisy real-world audio, distinguishing pauses from the end of sentences, and tracking speakers in meetings.

@alexandr_wang on X published this on Nov 10, 2026 (source: x.com).

Alexander Wang said on 10 November 2026 (17:03 UTC) that Muse now has its own email address, so paid subscribers in the United States and Canada can forward mail or add it to a thread and have the assistant act on it.

@mustafasuleyman on X published this on Nov 10, 2026 (source: x.com).

Microsoft AI CEO Mustafa Suleyman said on 10 November 2026 (17:01 UTC) that MAI-Transcribe-2 is available, calling it faster and cheaper than named alternatives and pricing it at $0.10 per hour of audio, with claims of being 10x faster than GPT-Transcribe and 5x faster than Gemini 3.5.

@Hailuo_AI on X published this on Nov 10, 2026 (source: x.com).

MiniMax's Hailuo account said on 10 November 2026 (16:56 UTC) that its image generation now supports up to 16 reference images and finishes in 10 seconds.

@nvidia on X published this on Nov 10, 2026 (source: x.com).

NVIDIA said on 10 November 2026 (16:08 UTC) that Alpamayo 2 Edge is available, bringing adaptive AI reasoning to the vehicle in real time, including planning driving paths and explaining decisions. The company said developers can build on the open model for their own fleets.

@ollama on X published this on Nov 9, 2026 (source: x.com).

Ollama said on 9 November 2026 (20:57 UTC) that a new usage page lets customers monitor daily and weekly usage, including spend per day, spend per API key, input and output tokens per model, and the cached-input rate. The original post is within 48 hours of this slot.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12282 on 11 November 2026 (published_at 2026-11-11T03:31:06Z, still 10 November in America/New_York). The notes say the chat path fixes Muse Spark ignoring response_format json_schema.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12277 on 11 November 2026 (published_at 2026-11-11T03:11:17Z, still 10 November in America/New_York). The notes say the mtmd path fixes a string format and renames it to mtmd_format.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12274 on 11 November 2026 (published_at 2026-11-11T02:52:55Z, still 10 November in America/New_York). The notes add FP8 KV-cache support, including tests and a path to enable Gemma 4 Assistant with FP8.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12273 on 11 November 2026 (published_at 2026-11-11T02:35:01Z, still 10 November in America/New_York). The notes merge a fix for sampling output aliasing.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12271 on 11 November 2026 (published_at 2026-11-11T02:14:59Z, still 10 November in America/New_York). The notes say the backend scheduler is reused when sched_params are compatible.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12270 on 11 November 2026 (published_at 2026-11-11T01:55:55Z, still 10 November in America/New_York). The notes add MUL_MAT src1 precision tests and gating in ggml.

Ggml-org published this on Nov 10, 2026 (source: github.com).

ggml-org published llama.cpp b12263 on 11 November 2026 (published_at 2026-11-11T01:26:31Z, still 10 November in America/New_York). The notes say the CUDA backend halves the VMM pool reservation size and retries when cuMemAddressReserve fails.

Microsoft published this on Nov 10, 2026 (source: techcommunity.microsoft.com).

Microsoft said on 10 November 2026 (published 2026-11-10T14:00:00Z) that multilingual post-stream refinement in Azure Speech, part of Microsoft Foundry, is generally available. The post says a live session can recognize multiple languages, including mid-conversation switches, and that the path is faster than the July public preview.

AWS published this on Nov 10, 2026 (source: aws.amazon.com).

AWS said on 10 November 2026 (published 2026-11-10T17:07:19-08:00) that Moonshot AI's Kimi K3.1 Pro is now available on Amazon Bedrock, including through the OpenAI-compatible Chat Completions, Responses, and Converse APIs.

NVIDIA Blog published this on Nov 10, 2026 (source: blogs.nvidia.com).

NVIDIA's blog on 11 November 2026 (published_at 2026-11-11T03:00:20+00:00, still 10 November in America/New_York) describes how Inception startups use vision AI and robotics to recover materials from mixed waste, with comments from NVIDIA's head of sustainability, Josh Parker.

Amazon (Kiro) published this on Nov 10, 2026 (source: kiro.dev).

Kiro's changelog dated 10 November 2026 says Enterprise Management is in preview for newly created profiles in US East (N. Virginia). Administrators write one policy in the AWS console, and Kiro applies it in the IDE (1.4.154 or later), CLI (3.1.0 or later), Kiro Web, and Kiro Crew LTS.