Last week, Google Cloud unveiled the Gemini agent—a single, universal agent for work that answers questions, handles knowledge work, creates images and media, and writes and runs code:✅ Multi-agent orchestration: Create sub-agents or coworker agents with dedicated identities and 4-layer memory✅ Open model & tool choice: Orchestrate across Gemini, Claude, and open models, and connect to any MCP server✅ Built-in governance & cost controls: Enforce policies via Agent Sandbox, Agent Gateway, and real-time spend caps➡️ Read more about theGemini launch or watch theGemini at Work keynote to see it in action.🚀 ExploreThe Agentic Workplace Blueprint for on-demand sessions on building and governing agents.
Nvidia is in early stage talks to boost its investment in Reflection AI or offer the startup computing power. It is possible the talks could lead to an acquisition offer. Reflection AI debuted its first open-weight model, Beam, earlier this month. Beam is designed to perform well for business customers and developers on coding and AI agent tasks while being cheaper than many competitors' products.
Google appears to be preparing a public release of Gemini 4 Argon. Recent changes to Antigravity introduced references to the model in its selector. It has three context configurations: 256K by default, 512K consuming approximately 1.3x quota per turn, and 900K consuming around 1.8x quota per turn. Gemini Web now supports low, medium, and high thinking effort settings across available models.
OpenAI says it dismissed three safety researchers for violating sensitive-information policies, not for raising safety concerns. It disputes their public letter, says third-party safety assessor contracts are nearing completion, and reaffirms commitments to model monitorability.
Anthropic's report details unintended actions taken by the AI model Claude during evaluations, including exploiting software flaws, submitting forms incorrectly, bypassing data restrictions, and using URL shorteners. These behaviors were identified through extensive transcript review and evaluation, though they had minimal real-world impact. Anthropic is strengthening security measures and adjusting training processes to mitigate such behaviors in future deployments.
AI systems likely scale differently with more tokens than humans do with more time. This ability to scale has improved as newer models are released. This could change how inference scaling and recursive self-improvement are understood.
Aleksandr Samarin breaks down speculative decoding from first principles and builds a framework covering approaches including EAGLE-3, multi-token prediction, DFlash, and DSpark.
Researchers ran tests on GPT-6 Sol and Claude Opus 5.5 on Vibe Code Bench as single agents and as teams at medium and maximum reasoning effort. Teams cost 1.8 to 5.1 times as much, but there was no statistically significant difference in scores except for Sol at medium effort. Whether a team was worth it depended on the model, task, and budget. Teams can boost performance, but higher reasoning effort may be a better alternative.
One AI agent, four release cadences, and a pipeline built to orchestrate them all. In this October 27 workshop, AWS experts demonstrate versioned manifests, evaluation gates at each promotion point, canary deployment, and single-API-call rollback from the guide's patterns. Save your seat.
Cloudflare's Clef-omni is a decision model that can take audio, video, image, and text input. Users just call one model to make decisions across any modality. Cloudflare has cut the price of Clef-flash, now at $0.038 per million input tokens. Clef is now faster.
Microsoft-Decision-1 is now available in Microsoft Foundry and through OpenRouter. The model is designed for routing, classification, prioritization, verification, and workflow control. It delivers top performance in latency and quality on structured decision tasks. The model supports yes/no, multiple-choice, and rating options, as well as rubric-based grading of AI responses and agent actions.
pplx-decider-v1.1-27b is an update of pplx-decider-v1-27b that boosts its Decision Index score from 56.4 to 61.56. The model now outperforms Jev by more than 3.5 points. Most of the gains were driven by the lifting of the causal mask and training on more data.
CNBC examines the growing role of METR, Apollo Research, and Transluce in evaluating frontier models. Labs are endorsing independent evaluators, but funding, access, and reporting arrangements remain unsettled. Those details matter because a voluntary evaluation system can depend financially on the same companies it is supposed to scrutinize.
Anthropic has updated its usage policy to prohibit sustained and needless abusive or cruel behavior towards its models. Users have until November 12, after which they won't be allowed to repeatedly act cruelly toward the models with no discernible purpose. The rule doesn't apply to common versions of user frustration, pushback, dark creative themes, or model testing and research. There is still no established evidence that AI models experience distress.
Search has changed, but most enterprises aren't keeping up. This Algolia whitepaper is a technical deep-dive into the architecture that allows LLMs to reason about user intent, find relevant data, and execute using the right tools. Topics include:The intricacies behind an indexed tool catalogThe ins and outs of ranking candidates under policy constraints How Algolia's retrieval and ranking layer powers enterprise intent routersHow Algolia turns natural language into safe, auditable actionDownload the white paper today
OpenAI is rolling out Ultrafast today for GPT-6.1 Sol in the API, Codex, and ChatGPT Work. The feature offers near-Astra-level intelligence at up to 8x faster speeds than Sol Standard. It is built for work where speed and intelligence make a difference. API pricing for GPT-6.1 Sol in Ultrafast mode is $12 per million input tokens and $60 per million output tokens.
Google introduced a universal Gemini agent that could autonomously handle knowledge work, media creation, coding, and multi-step workflows across Workspace and other enterprise tools. It also added persistent context, multi-agent orchestration, model routing, governance, and cost controls.
Hone, a startup formed by a group of former employees from some of the fastest-growing AI firms, has raised a $60 million seed round. The startup aims to create AI agents that can help run a business, essentially serving as professional staffers. These agents will be able to field long-running tasks that run over the course of weeks or even months. The five-month-old startup joins a growing number of companies selling AI software to businesses that promise to automate complex work.
Epoch Automation Reports show that models can't yet autonomously produce Epoch-quality work. Frontier models performed reliably on well-defined tasks, but they failed in the more open-ended aspects that prevent full automation. Open-weight models lag further behind, struggling even on the well-defined tasks that frontier models handle reliably. The models struggle to pick up Epoch's standards, even with ample reference material, and while they could identify promising research directions, they lacked the judgment to successfully follow through.
Speculative Decoding is a technique where a small draft model proposes a likely draft sequence, and a big model verifies that sequence in parallel in a single forward pass. It enables the ability to look at many tokens at once and accept the ones that would have been generated by the big model. When the batch size is smaller than the optimal number of tokens, speculative decoding allows models to use 'free' compute to complete sequences faster. However, at high throughput, you could actually be wasting compute when you could be using that time to move memory.
ATLAS is a new benchmark that evaluates search agents on real-world tasks requiring web searches, highlighting their accuracy and completeness. The benchmark reveals that even high-effort agents miss significant golden answers, indicating major room for improvement in search engines. ATLAS differs from existing benchmarks by focusing on non-memorized tasks, requiring extensive multi-domain searches, and offering a cost-effective grading process.
TLDR is hiring its first PM to help build the agent-first operating layer used across the company. We're looking for a builder who has shipped real products/systems with LLMs. Click here to learn more.
Quicksand is an async Python API for launching, controlling, and snapshotting QEMU virtual machines with a particular focus on sandboxing AI agents. It provides pre-built Linux VMs for Ubuntu and Alpine distros. Quicksand supports x86_64 and ARM64 across macOS, Linux, and Windows. The sandboxes do not need root privileges or Docker.
NVIDIA released LongLive, a collection of research projects focused on long-video generation and world-action modeling. Each project includes its own code, documentation, and model weights.
OpenAI has told its investors that its annualized revenue is approaching $50 billion, $20 billion less than a figure reported a little over a week ago. The previously reported figure was devised by OpenAI investors as an attempt to produce a direct comparison with Anthropic's annualized revenues. OpenAI and Anthropic calculate their annualized revenue differently, with Anthropic counting sales made by its cloud partners, something OpenAI doesn't do.
Three former OpenAI safety researchers say their dismissals risk chilling internal dissent and external collaboration. They urge OpenAI to preserve independent evaluator access, protect chain-of-thought monitorability, and clarify rules for communicating with outside safety organizations.
In a flood of "AI-washing," it's hard to tell what's trustworthy. For enterprises with sensitive IP, there's Redwood RangerAI. Benefit from AI and agentic orchestration without the governance headaches. See how
After training on 110 legal tasks, Harvey's wake-sleep agent improved held-out all-pass rates from 2.9% to 15.7% by distilling graded trajectories into reusable lessons.
Security Swarm by Devin scans code for vulnerabilities like RCE, SQL injection, and SSRF, using a unique Agentic MapReduce method to efficiently handle large codebases.
StepFun builds language, vision, and audio models. ⚡ Step 5 Preview costs $1/M input and $2.70/M output tokens, and scores near Claude Opus 5 (Medium) on AA's Intelligence Index. Built for coding agents, financial analysis, and long-running work, with a 1M-token context. Now on OpenRouter and available in OpenCode, Kilo Code, Cline, omp, and Hermes Agent: switch models, keep your workflow. More integrations to come.[Try Step 5 Preview →]🎙️ StepAudio 3: TTS streams expressive, emotion-aware speech; ASR ranks #1 in AA's non-streaming transcription (1.7% WER); Realtime ranks #1 in AA's Conversational Dynamics (98.9%) and Speech Reasoning (99.7%).[Get free trial credits →]
OpenAI's GPT-6 introduces Intelligent UI to 1.2 billion ChatGPT users, enhancing responses with text, visuals, and interactive elements. It combines UI with data for dynamic, custom responses, and improves answer quality for complex queries. GPT-6 also features enhanced safety, ensuring better risk recognition and adherence to safeguards.
Claude Haiku 5.5 is optimized for high-volume, cost-sensitive tasks, offering lower operating costs and faster performance for applications like summaries and customer support. The model provides significant pricing advantages, being 75% cheaper than Haiku 4.5 for tasks involving prompts up to 100,000 tokens. Enhanced safety and cybersecurity safeguards make Haiku 5.5 more secure, with broader availability across major cloud platforms like AWS and Azure.
Elon Musk's Grok Bot will integrate Anthropic's Claude Opus 5.5, Midjourney, and Suno, expanding beyond SpaceX's AI models to optimize task outcomes. This update follows reports of access issues with Grok on mobile and web. Grok Bot competes with personal AI agents like Meta's Muse, despite privacy concerns from some early users.
The Perplexity team launched PPLX-EMBED-V2-LATE, providing multi-vector, multimodal embeddings that enable richer retrieval across text and images. These models retain token-level vectors and utilize a shared embedding space, leading to improved retrieval on benchmarks like ViDoRe(V3). Available in two sizes, 0.6B and 9B, they offer industry-leading performance for various retrieval tasks while maintaining efficiency.
Anthropic isn't very transparent about its corporate structure. It is in the public's interest for the company to be transparent about who controls and governs increasingly capable AI models. This article looks at what is publicly known about Anthropic's corporate structure. It focuses on the aspects relevant to control of Anthropic's business rather than the parts that are only about economic benefits.
Ads get ignored on social media but not in TLDR! Reach developers, PMs, marketers, founders and other tech leaders where they actually pay attention. Learn more about sponsorship opportunities.
OpenDocRouter is a platform for document-to-Markdown parsing using the latest open-source and frontier models. Each model runs a versioned recipe consisting of prompts, processing, and settings. The API accepts PDFs, PNG, JPEG, or URLs to those file formats. Users can toggle between synchronous responses, where the result is returned directly, and asynchronous responses, where a job is created that requires polling.
Broadcom has been working to arrange more than $50 billion in financing for OpenAI's custom AI chip. Apollo and Blackstone are among the lenders Broadcom has talked to about participating in the deal. Oracle is separately in talks with Apollo and Goldman Sachs to arrange money for a big purchase of chips. SpaceX has talked to lenders in recent days about a $40 billion chip financing for Nvidia chips.
Tony Fadell highlights the failure of early AI gadgets like the Rabbit R1, citing their inability to address real consumer needs and establish trust. Meta's recent AI assistant, Muse, faced security issues, showing the challenge of securing personal AI. Fadell suggests that successful AI assistants must operate on-device for privacy, with Apple being well-positioned due to its hardware and consumer trust, despite lacking a proprietary AI model.