xAI just rolled out Grok 4.3 alongside a new web-based voice cloning suite. The flagship model features permanently active reasoning, a 1-million-token context window, and agentic tool use like web browsing and code execution. API pricing has dropped roughly 40% to 60% compared to Grok 4.2, costing $1.25 per million input tokens and $2.50 per million output tokens.
Microsoft just rolled out Legal Agent in Word on Windows for Frontier program members worldwide. Designed for contract review and negotiation, the agent compares clauses against company playbooks and inserts precise, tracked-change redlines without messing up formatting. Why it matters: Instead of letting an LLM blindly rewrite legal text, it uses deterministic logic to preserve negotiation history and structure, taking the grunt work off legal teams.
OpenAI's GPT-5.5 just became the second AI model to fully crack UK AISI's enterprise cyberattack simulation, trailing only Anthropic's Claude Mythos. It completed the 32-step penetration test—typically a 20-hour job for human experts—in 2 out of 10 attempts while hitting a 71.4% pass rate on expert-level CTF challenges, though industrial control systems remain unbreached.
Perplexity just rolled out major upgrades to its enterprise agent Computer, embedding it directly into Microsoft Teams and Excel. Users can now summon the assistant in chats, query financial models from a native spreadsheet sidebar, and run more than 70 prebuilt workflows. A new 1Password integration also keeps credentials secure while acting on the user's behalf. Why it matters: Perplexity is pushing deep into everyday office surfaces to challenge Copilot on its home turf.
Cursor just launched a public beta TypeScript SDK that lets developers build programmatic coding agents using the exact same runtime and harness powering its IDE. It runs locally, on self-hosted workers, or inside Cursor's dedicated cloud VMs with support for any frontier model, including Composer 2.5. Why it matters: teams can now embed headless coding agents straight into CI/CD pipelines and internal tools without rebuilding the agent harness from scratch.
Anthropic just integrated its claude-api skill into CodeRabbit, JetBrains, Resolve AI, and Warp. Originally launched in Claude Code, the tool helps developers automatically handle prompt caching, update agent patterns, and migrate code to new models like Claude Opus 4.7 directly within their IDEs. The skill is also open source on GitHub.
Researchers just introduced RecursiveMAS, a framework where AI agents collaborate directly within latent representations instead of passing text back and forth. By keeping intermediate thoughts silent until the final answer, it slashes token usage by up to 75% while boosting inference speeds up to 2.4x. It essentially swaps slow text decoding for seamless shared reasoning.
NVIDIA just unveiled Nemotron 3 Nano Omni, an open multimodal model unifying vision, audio, and language into a single 30B-A3B architecture. It delivers up to 9x higher throughput than comparable open omni models by eliminating the need to pass data across separate perception models. Why it matters: Agentic workflows like GUI navigation and document processing can now run in real time with far lower latency.
Snap just launched AI Sponsored Snaps, a new ad format that places brand AI agents directly inside Snapchat's Chat inbox. Users can ask questions and get recommendations in real time without leaving the app, with credit bureau Experian signing on as the first alpha partner. With nearly a billion monthly users, Snap is betting the future of ads looks less like banners and more like texting a bot.
Poolside just launched its first public coding models, headlined by Laguna XS.2, an open-weights 33B model (3B active) under Apache 2.0 that runs on a single GPU. The lab also unveiled Laguna M.1, a massive 225B MoE, alongside preview tools pool (a terminal agent) and Shimmer (a cloud sandbox). Both models are temporarily free via API and OpenRouter, bringing serious agentic horsepower to local setups.
OpenAI and AWS just expanded their strategic partnership, rolling out a limited preview of GPT-5.5, Codex, and OpenAI-powered Managed Agents on Amazon Bedrock. Enterprise developers can now run OpenAI's frontier models and coding workflows directly inside their existing AWS security and governance perimeter, with Codex usage counting toward AWS commitments.
Researchers just introduced OneManCompany, a framework that ditches fixed agent org charts in favor of a talent market where specialized agents get recruited dynamically per task. Using an Explore-Execute-Review tree search and automated HR and COO roles, it scored an 84.67% success rate on PRDBench. Why it matters: treating agents as a recruitable workforce keeps open-ended workflows from breaking when tasks evolve.
Cognition just launched Devin for Terminal, bringing its autonomous coding agent directly into your local shell. It handles development tasks locally without requiring cloud infrastructure, but lets you seamlessly hand off heavy workloads to the cloud when your laptop hits its limits.
Researchers just introduced Agent-Native Research Artifacts (ARA), a protocol designed to replace traditional PDF papers with executable, machine-readable packages. ARA preserves the dead ends and unwritten engineering details that linear papers discard, boosting AI reproduction rates from 57.4% to 64.4%. Why it matters: as agents become science's primary readers, publication is moving from human storytelling to executable knowledge.
Microsoft just rolled out real-time voice agents to general availability in Copilot Studio, replacing rigid phone trees with natural, interruptible voice AI. The rollout brings new autonomous agents to Dynamics 365 Contact Center and Sales for support, QA, and deal tracking. It effectively shifts CRM from manual data entry to proactive AI automation.
Sakana AI just unveiled Conductor, a 7B model trained not to solve tasks directly, but to act as a manager orchestrating frontier models like GPT-5 and Claude. It writes dynamic prompts and delegates subtasks in natural language, beating every individual model in its pool on LiveCodeBench at a fraction of typical multi-agent costs.
OpenAI is working on its very own AI-first smartphone aimed at taking on the iPhone, according to analyst Ming-Chi Kuo. The company is reportedly partnering with Qualcomm and MediaTek on processors and tapping Luxshare for manufacturing, targeting mass production by 2028. Why it matters: Rather than replacing phones, OpenAI seems to think agentic AI will redefine them.
OpenAI just open-sourced Symphony, a spec that turns issue trackers like Linear into an orchestration engine for Codex coding agents. Instead of engineers juggling multiple interactive terminal sessions, every open ticket automatically gets a dedicated agent that runs until completion. OpenAI reported a 500% jump in landed PRs on internal teams, showing how delegating whole tasks frees human attention.
Chinese regulators just ordered Meta to walk away from its planned $2B acquisition of AI agent startup Manus. Authorities prohibited foreign investment in the deal and demanded both sides withdraw. Manus builds agents that can plan and complete tasks independently. Why it matters: Meta had already begun deeply integrating the team, making any forced unwinding an operational and legal nightmare.
Meta just announced a massive deal with AWS to deploy tens of millions of custom Graviton CPU cores to power its agentic AI systems. Autonomous agents require substantial CPU horsepower to reason and execute complex tasks, and AWS Graviton5 chips bring faster processing and higher bandwidth. The move makes Meta one of AWS's largest Graviton customers as it aggressively diversifies its compute fleet.
