AI Daily | 2026-09-26

Model Releases & Updates

LFM2.5-VL-3B-DSpark LLiquid AI

• Bottom line：Liquid AI has released LFM2.5-VL-3B-DSpark, an experimental speculative decoding draft model designed to accelerate vision-language model inference.
• Core Highlights：Adds approximately 279.5M parameters (an 8.9% increase) specifically tailored as a speculative decoding draft model for LFM2.5-VL-3B.
• Performance：Achieves up to 3.13x faster decoding speed on Apple Silicon and up to 2.66x faster decoding on NVIDIA H100 hardware.
• Source：MarkTechPost Coverage

Product Releases & Updates

Turnstile Spin CCloudflare

• Bottom line：Cloudflare has introduced Turnstile Spin, enabling autonomous AI agents to set up and manage end-to-end web security challenges.
• Core Highlights：Designed as an agent-mediated end-to-end implementation of Cloudflare’s privacy-first client-side challenge system.
• Implementation：Requires a straightforward two-step process involving frontend widget rendering and backend Siteverify API token validation.
• Source：Cloudflare Blog

Runway Layers RRunway

• Bottom line：Runway has launched Layers, a new editing capability that splits any image into editable components with a single click.
• Core Highlights：Allows users to instantly remove backgrounds, edit text, and rework individual elements independently within the Runway platform.
• Workflow Impact：Streamlines creative editing by eliminating the need to export images to external tools for basic layer segmentation.
• Source：

  View post on X by @runwayml

  在 X 上查看 @runwayml 的貼文 ↗

Memorystore for Valkey 9.1 GGoogle Cloud

• Bottom line：Google Cloud has announced the general availability of Memorystore for Valkey 9.1, delivering massive performance boosts for AI and microservice backends.
• Core Highlights：Achieves up to 3x higher queries per second (QPS) while maintaining microsecond latency compared to older Redis Cluster implementations.
• Ecosystem Context：Built on the open-source Valkey alternative governed by the Linux Foundation to support high-throughput cloud architectures.
• Source：Google Cloud Blog

Industry News

Cognition Crosses $1B Annualized Revenue Run Rate CCognition

• Bottom line：AI software development pioneer Cognition has surpassed $1 billion in annualized revenue run rate less than two years after launching Devin.
• Core Highlights：Reflects rapid enterprise adoption across engineering teams at major companies like GE Aerospace, Rivian, and Exa.
• Milestone Significance：Highlights the massive commercial scaling of autonomous coding agents and developer platforms.
• Source：Cognition Blog

Anthropic Founders Seek Voting Control Ahead of IPO AAnthropic

• Bottom line：Anthropic is reportedly asking shareholders to approve a new governance structure that grants its co-founders majority voting control.
• Core Highlights：CEO Dario Amodei and six co-founders would hold 50.1% of voting power over corporate matters under the proposed arrangement.
• Strategic Context：Positions the safety-focused AI lab similarly to other tech firms establishing special voting shares ahead of public market debuts.
• Source：TechCrunch Report

Appellate Court Rules on Pentagon Supply Chain Risk for Anthropic UUS Federal Appeals Court

• Bottom line：A federal appeals court has upheld the Department of Defense’s authority to designate Anthropic as a supply chain risk over safety restrictions.
• Core Highlights：The 2-to-1 ruling supports excluding Anthropic and Claude from military contracts due to built-in safety limits triggering operational uncertainties.
• Legal Precedent：Reinforces government oversight regarding frontier AI safety protocols and defense procurement standards.
• Source：Ars Technica Coverage

Research Papers

Taste-Bench: Evaluating Agent Taste and Decision Forks MMicrosoft Research & Collaborators

• Bottom line：Researchers have introduced Taste-Bench to evaluate how frequently frontier AI agents choose optimal decision paths during long-horizon software engineering tasks.
• Core Highlights：Mines decision forks automatically from parallel engineering and research trajectories where specific choices lead to better outcomes.
• Key Findings：Shows that current top frontier models select the superior direction less than 60% of the time, and increased reasoning budgets do not automatically raise accuracy.
• Source：arXiv Paper

Other Highlights

GitHub Copilot App Canvases Tutorial GGitHub

• Bottom line：GitHub has published an official guide detailing how developers can use natural language descriptions to generate custom interactive canvases.
• Core Highlights：Introduces the /create-canvas skill to build customizable user interfaces directly within the GitHub Copilot app environment.
• Practical Application：Aims to help developers rapidly prototype and structure custom developer workflows without manual scaffolding.
• Source：GitHub Blog