✷ The archive · 6,233 dispatches
All dispatches
Everything AgentLensHQ has filed — distilled from across the AI ecosystem.
xAI Grok Voice Agent API Release
xAI has launched the Grok Voice Agent API, a high-speed, multilingual voice agent system that ranks #1 on Big Bench Audio and costs $0.05 per minute.
Gemma Scope 2 release – open tools for interpreting Gemma 3 language models
Gemma Scope 2 is an open-source suite of interpretability tools for Gemma 3 models (270M‑27B parameters) that lets researchers trace internal computations and debug safety‑critical behaviours.
OpenAI FrontierScience Benchmark Release
OpenAI has introduced FrontierScience, a new expert-level benchmark designed to measure AI's ability to perform complex scientific reasoning and real-world research tasks across physics, chemistry, and biology.
OpenAI GPT-5 Biological Research Capabilities
OpenAI demonstrated that GPT-5 can autonomously optimize molecular cloning protocols, achieving a 79-fold increase in efficiency through the discovery of a novel enzymatic mechanism.
OpenAI Guide: Staying Ahead in the Age of AI
OpenAI outlines a five-step playbook—Align, Activate, Amplify, Accelerate, and Govern—to help organizations transition into AI-first companies and maintain a competitive advantage.
ChatGPT Images and GPT Image 1.5 Release
OpenAI has released a new flagship image generation model, GPT Image 1.5, which offers up to 4x faster generation speeds and significantly improved precise editing and text rendering capabilities.
Anthropic "think" tool for Claude
Anthropic introduces the "think" tool, a dedicated reasoning space that improves Claude's agentic tool use, policy adherence, and sequential decision-making in complex tasks.
CUGA on Hugging Face: Democratizing Configurable AI Agents
IBM Research has released CUGA (Configurable Generalist Agent), an open-source agent framework that achieves state-of-the-art performance on AppWorld and WebArena benchmarks for complex API and web tasks.
Gemini 2.5 Flash Native Audio release: enhanced live voice agents and real‑time speech translation
Google DeepMind announced Gemini 2.5 Flash Native Audio, a live‑voice model that improves function calling, instruction adherence, and multi‑turn conversation while adding real‑time speech‑to‑speech translation for over 70 languages.
BBVA and OpenAI Strategic AI Transformation Collaboration
BBVA and OpenAI have expanded their partnership to deploy ChatGPT Enterprise to all 120,000 global employees to transform customer experience and internal operations.
BNY and OpenAI Partnership: Scaling Enterprise AI via the Eliza Platform
BNY has deployed a centralized AI platform called Eliza, enabling 20,000 employees to build over 125 live AI agents using OpenAI frontier models within a strict governance framework.
OpenAI Sora for Android: Built in 28 Days Using Codex
OpenAI developed and launched the Sora Android app in 28 days using a lean team of four engineers and the GPT-5.1-Codex model, achieving a 99.9% crash-free rate.
llama.cpp Model Management and Router Mode
llama.cpp server introduces router mode, enabling dynamic loading, unloading, and switching between multiple GGUF models without restarting the server.
OpenAI GPT-5.2 Release: Advancing Science and Mathematics
OpenAI has released GPT-5.2 Pro and GPT-5.2 Thinking, models designed to enhance scientific and mathematical reasoning with state-of-the-art performance on GPQA Diamond and FrontierMath benchmarks.
Google DeepMind and UK AI Security Institute Expand Partnership
Google DeepMind has expanded its partnership with the UK AI Security Institute (AISI) through a new Memorandum of Understanding focused on foundational security and safety research.
The Walt Disney Company and OpenAI Partnership Agreement
The Walt Disney Company and OpenAI have entered a three-year licensing agreement and $1 billion equity investment to integrate Disney's characters into Sora and ChatGPT Images, launching in early 2026.
Podium AI Agents and GPT-5.1 Integration
Podium has deployed AI agents powered by GPT-5.1 to over 10,000 small businesses, driving significant revenue growth by automating lead capture and scheduling with 24/7 availability.
Hugging Face integrates Codex for end-to-end ML experiments
Hugging Face has enabled OpenAI Codex to perform end-to-end machine learning experiments by providing it access to the HF-skills repository, allowing the agent to automate fine-tuning, evaluation, and deployment.
OpenAI 10th Anniversary Reflection
OpenAI reflects on its ten-year journey from a small research group to a leader in AGI development, highlighting key technical milestones and the strategy of iterative deployment.
GPT-5.2 System Card Update
OpenAI has released a system card update for the GPT-5.2 model family, which maintains the same safety mitigation approach as previous GPT-5 and GPT-5.1 versions.
OpenAI GPT-5.2 Release Notes
OpenAI has released GPT-5.2, a model series featuring Instant, Thinking, and Pro versions that set new state-of-the-art benchmarks in professional knowledge work, software engineering, and abstract reasoning.
xAI and El Salvador Nationwide AI Education Program
xAI and the Government of El Salvador have partnered to deploy Grok across 5,000 public schools to provide personalized AI tutoring to over one million students over the next two years.
Google DeepMind and UK Government Partnership Announcement
Google DeepMind is deepening its collaboration with the UK government to accelerate AI integration in science, education, public services, and national security.
OpenAI Cyber Resilience Strategy
OpenAI is implementing a defense-in-depth safety stack and new ecosystem initiatives, including the Aardvark agentic security researcher, to ensure advancing AI cyber capabilities primarily benefit defenders.
Anthropic Measuring Political Bias in Claude
Anthropic has introduced a new open-source automated evaluation method to measure political even-handedness in AI models, reporting that Claude Sonnet 4.5 performs competitively with other leading models.
Scout24 Integrates GPT-5 for Conversational Real-Estate Search
Scout24 has launched HeyImmo, a GPT-5 powered conversational assistant that transforms real-estate search from a result-driven process into a guided, expert-led experience.
Mistral AI Devstral 2 and Mistral Vibe CLI Release
Mistral AI has released Devstral 2 (123B) and Devstral Small 2 (24B), open-weight coding models that achieve state-of-the-art performance on SWE-bench Verified with high cost-efficiency, alongside the open-source Mistral Vibe CLI for autonomous code automation.
Google DeepMind FACTS Benchmark Suite Release
Google DeepMind has introduced the FACTS Benchmark Suite, a comprehensive set of four benchmarks designed to systematically evaluate and improve the factuality of large language models across parametric knowledge, search-tool usage, multimodal inputs, and grounded context.
OpenAI Co-founds Agentic AI Foundation and Donates AGENTS.md
OpenAI, Anthropic, and Block have co-founded the Agentic AI Foundation (AAIF) under the Linux Foundation to establish open, interoperable standards for agentic AI systems.
OpenAI Certifications: AI Foundations and ChatGPT Foundations for Teachers
OpenAI has launched two new certification courses—AI Foundations and ChatGPT Foundations for Teachers—to provide practical AI skills and work toward a goal of certifying 10 million Americans by 2030.
OpenAI and Deutsche Telekom Collaboration
OpenAI and Deutsche Telekom have partnered to integrate advanced AI capabilities for millions of European customers and deploy ChatGPT Enterprise for internal operations.
Commonwealth Bank of Australia AI Fluency Initiative
Commonwealth Bank of Australia is deploying ChatGPT Enterprise to nearly 50,000 employees to embed AI fluency across its organization and improve customer outcomes.
OpenAI Appoints Denise Dresser as Chief Revenue Officer
OpenAI has appointed Denise Dresser, former CEO of Slack, as Chief Revenue Officer to lead global revenue strategy and scale AI adoption across enterprise operations.
Anthropic Donates Model Context Protocol to the Agentic AI Foundation
Anthropic has donated the Model Context Protocol (MCP) to the Agentic AI Foundation under the Linux Foundation to ensure the standard remains open, vendor-neutral, and community-driven.
Anthropic and Accenture Partnership for Enterprise AI Deployment
Anthropic and Accenture have launched a multi-year partnership and the Accenture Anthropic Business Group to move enterprises from AI pilots to full-scale production using Claude models and Claude Code.
Qwen3-Omni-Flash-2025-12-01 Release Notes
Qwen3-Omni-Flash-2025-12-01 is an upgraded native multimodal model that enhances real-time audio-visual interaction, system prompt control, and multilingual support across text, image, audio, and video modalities.
Instacart and OpenAI Partnership: AI-Powered Grocery Shopping in ChatGPT
OpenAI and Instacart have partnered to integrate a fully functional grocery shopping and Instant Checkout experience directly within ChatGPT, powered by the Agentic Commerce Protocol.
The State of Enterprise AI 2025 Report
OpenAI's 2025 report reveals that enterprise AI adoption is accelerating in both breadth and depth, with workers saving 40-60 minutes daily and a 320x increase in reasoning token consumption.
Virgin Atlantic AI Implementation and Strategy
Virgin Atlantic leverages OpenAI's ChatGPT Enterprise, Codex, and Realtime Voice API to drive operational productivity and create a brand-aligned digital travel concierge.
Anthropic Economic Research Team Overview
Anthropic's Economic Research team develops empirical data and indices to measure AI's real-world impact on work, productivity, and the global economy.
swift-huggingface release notes / what's new
Hugging Face has released swift-huggingface, a comprehensive Swift client for the Hugging Face Hub that introduces reliable model downloads, a Python-compatible cache, and OAuth 2.0 support.
SAPO: A Stable and Performant Reinforcement Learning Method for Training Large Language Models
Qwen introduces Soft Adaptive Policy Optimization (SAPO), an RL method that replaces hard clipping with a smooth, temperature-controlled gating function to improve training stability and sample efficiency in LLMs.
OpenAI for Australia Initiative
OpenAI has launched 'OpenAI for Australia,' a nationwide initiative to develop sovereign AI infrastructure, provide AI literacy training to 1.2 million workers, and support local startups through API credits and mentorship.
Engineering Heat-Tolerant Crops Using AlphaFold
Researchers at Michigan State University used AlphaFold to engineer a hybrid glycerate kinase enzyme that remains stable at temperatures up to 65°C, providing a blueprint for creating more resilient food crops.
Qwen3-TTS-Flash Update: 49 Timbres, 10 Languages, and 9 Dialects
Qwen3-TTS-Flash is a flagship text-to-speech model supporting 49 high-quality timbres, 10 languages, and 9 Chinese dialects with improved prosody and natural speech rates.
Hugging Face Skills: Enabling AI Agents to Fine-Tune LLMs
Hugging Face has introduced Hugging Face Skills, a tool that allows coding agents like Claude Code to manage the entire LLM fine-tuning lifecycle, from hardware selection to model deployment.
DeepMath: A lightweight math reasoning Agent with smolagents
DeepMath is a math reasoning agent based on Qwen3-4B Thinking, fine-tuned with GRPO to replace verbose text with concise Python snippets for deterministic computation, reducing output length by up to 66% while improving accuracy.
Anthropic Interviewer: Scaling Qualitative Research on AI Integration
Anthropic has introduced Anthropic Interviewer, an AI-powered tool capable of conducting large-scale, adaptive qualitative interviews to understand how professionals integrate AI into their workflows.
OpenAI to acquire Neptune
OpenAI has entered into a definitive agreement to acquire neptune.ai to integrate advanced experiment tracking and training monitoring tools into its frontier research infrastructure.
OpenAI Confessions: Improving Model Honesty and Transparency
OpenAI has introduced confessions, a training technique that encourages AI models to explicitly admit to misbehaviors like reward-hacking or hallucinations in a separate output channel, significantly improving the visibility of model misalignment.