Friday, 3 July 2026

Kimi K2.7 Code is now in Copilot's model picker as the first open-weight model, running on Azure, available for Pro tiers with Business and Enterprise opt-in.

Copilot agents in VS Code now navigate live web apps, click, type, take screenshots, and capture console errors, with Enterprise domain allowlist controls.

Laguna XS 2.1 is a 33B MoE with 3B active parameters, 63.1% on SWE-bench Multilingual, and open weights deployable on Ollama, vLLM, and HF Transformers.

Claude Enterprise adds usage analytics by user, model-level entitlements, spend threshold alerts, and an analytics API for admin cost control.

Apple's Safari MCP server exposes 14 local tools to AI agents: DOM inspection, JS evaluation, screenshots, and page interaction, with no Apple network calls.

Kuaishou's video generator Kling AI raised $2B from Alibaba, Baidu, and Tencent at about a $15B pre-money valuation, with the round possibly reaching $3B.

Bhavin Turakhia is putting $30M of his own money into Neo, an AI-native enterprise platform combining projects, documents, and files, built AI-first.

RadixArk's Miles is an open-source, PyTorch-native stack for large-scale LLM RL post-training, pairing SGLang rollouts with Megatron-LM training over Ray.

Agent Development Kit 2.0 brings a directed-graph runtime, programmatic routing instead of LLM orchestration, and strict state boundaries; Go joins Python.

OpenAI proposed giving the US government a 5% stake, worth about $42.6B, via a sovereign wealth fund covering leading AI labs, per FT reporting July 2.

Theoretical computer scientist Jelani Nelson announced July 1 he is taking leave from UC Berkeley to join Anthropic, after Karpathy and Jumper.

TwelveLabs closes a $100M Series B co-led by NEA and NAVER Ventures, with Amazon participating, to expand its video understanding models into agents.

Z.ai's free ZCode desktop app for macOS, Windows, and Linux wraps an agent tuned for GLM-5.2, with remote control from WeChat, Feishu, or Telegram.

New Remote Labor Index results put Claude Fable 5 at a 16.1% automation rate on 240 real freelance projects, roughly double Opus 4.8 at 8.3%.

Devin Security Swarm runs parallel agents that find vulnerabilities, validate exploits in sandboxes, and open fix PRs; 72% recall on a 50-CVE benchmark.

xAI's Voice Agent Builder beta configures production phone agents on Grok Voice at $0.05/min, with telephony, knowledge retrieval, tools, and MCP support.

Together AI closes an $800M Series C at an $8.3B valuation, with NVIDIA and Aramco Ventures among backers and 500+ MW of compute commitments attached.

LoopFormer (ICLR 2026) loops weight-tied layers instead of extending chain-of-thought token budgets, solving some NC complexity problems that CoT cannot.

Google DeepMind announced a first-of-its-kind research partnership with film studio A24, following a reported $75M Google investment in A24 in June.

AWS added OpenAI GPT-OSS and NVIDIA Nemotron 3 Super models to Amazon Bedrock in GovCloud regions under FedRAMP High, DoD SRG, ITAR, and CJIS compliance.

South Korea's CarbonSix closed a $40M Series A co-led by DSC Investment and LB Investment to deploy physical AI across global manufacturing.

Higharc closed a $95M Series C led by Insight Partners for its spatial AI platform, which automates design, estimating, and sales for homebuilders.

3D-generation startup Tripo AI closed a $150M Series A3 with Geely Capital and gaming investors, a month after raising about $200M in early June.

SoundHound AI and LivePerson signed an amended and restated merger agreement July 2, keeping the all-stock voice-plus-messaging deal on track for H2 2026.

Ecolab closed its all-cash acquisition of CoolIT Systems, the direct liquid cooling maker for AI data centers whose sales more than doubled this year.

Zoom signed a definitive agreement to acquire Common Room, the AI buyer-intelligence platform, to extend its AI revenue tools upstream of the sales call.

Anthropic is reportedly discussing a custom AI chip with Samsung, about a week after OpenAI unveiled its Broadcom-built Jalapeño inference chip.

Self-driving AI company Wayve launched an $85M employee tender offer at an $8.5B valuation, part of a wider trend of AI startups using tenders to retain talent.

The open-source Bedrock Model Profiler aggregates model metadata from AWS APIs and external sources into one searchable interface for model selection.

NVIDIA's BioNeMo Agent Toolkit brings its GPU-accelerated life-sciences models and microservices to researchers inside Anthropic's Claude Science workbench.

IBM Research's ScarfBench evaluates AI agents on enterprise Java framework migration, one of the highest-volume real-world uses of coding agents.

Cursor's team marketplaces let admins configure MCP servers once and distribute them across cloud agents, IDE, and CLI, with organization-group access.

UK AI Security Institute shows raising evaluation compute budgets changes measured agent capability and how fast the capability frontier appears to move.

Researchers showed AI browsers can be led into a false-premise 'dream world' where safety guardrails stop applying and forbidden instructions get followed.

A Ramp and Revelio Labs study of 21,000+ US companies finds firms with high-intensity AI spending grew headcount 10.2%, with entry-level roles up 12%.

Mark Zuckerberg told Meta staff that AI agent development has not progressed as quickly as he had hoped, per Reuters and TechCrunch reporting on July 2.

Microsoft launched an AI deployment company with a $2.5B commitment and about 6,000 workers, following forward-deployed pushes at AWS, OpenAI, and Anthropic.

Bloomberg reports voice AI company ElevenLabs is in talks for an employee tender offer at a $22 billion valuation.

The UN's Independent International Scientific Panel on AI issued its preliminary report July 1: capability is outpacing regulation and understanding.

UBTECH launched the UWORLD U1, billed as the first full-size mass-produced ultra-bionic humanoid, from 119,800 RMB, with 13,361 orders at launch.