ai-agents

Anthropic's midsize model matches near-flagship agentic performance at a fraction of the price, and arrives as the lab lifts controls on Fable 5.

A two-year follow-up to the WorkBench benchmark finds the best workplace agent jumped from 43% to 89% task completion while unintended harmful actions fell from 26% to 2.5%.

The 2026-07-28 MCP release candidate makes the protocol core stateless, formalizes an Extensions framework, and deprecates Roots, Sampling, and Logging. The RC locked May 21 for a July 28 final.

Mistral rebranded its Le Chat assistant as Vibe, a single agent spanning Work Mode for office tasks and Code Mode for remote coding.

GitLab's next-generation source code management, shown at its Transcend event, keeps the Git protocol but rebuilds the backend so agents query repos server-side instead of cloning them.

An ICLR paper from MIT CSAIL and Harvard shows Monte Carlo inference helps Llama 4 Scout outpace GPT-5 at a Battleship test bed for around 1% of its cost.

At Build 2026, Microsoft revealed Project Solara, a chip-to-cloud platform built on AOSP for devices that run AI agents instead of apps, including a reference-design wearable badge.

Raindrop released Workshop, a free local debugger that streams an AI agent's tokens, tool calls, and spans to a browser and lets Claude Code write and fix evals against the trace.

At Build 2026 in San Francisco, Microsoft unveiled Project Polaris to replace GPT-4 Turbo in GitHub Copilot, open-sourced the Windows Agent Framework, and previewed the Windows Agent Runtime and Azure Agent Mesh.

Veeam used its VeeamON 2026 conference in New York City to preview v13.1 of its Data Platform with 70+ new features, a new DataAI Command Platform, and post-quantum cryptography support.

VS Code 1.121, released May 20, adds remote agent sessions over SSH and dev tunnels, native Mermaid diagram rendering, HTML file preview, and a new Agent Host Protocol for multi-client agent coordination.

Qwen3.7-Max debuts at Alibaba's Cloud Summit in Hangzhou as the company's most capable agent model, sustaining continuous autonomous execution for 35 hours in internal testing.