Skip to content
News6 min read14 March 2026

Claude's 1 million tokens of context is live. And your OpenClaw agent watches along in your browser.

Claude Opus and Sonnet get 1 million context tokens by default. OpenClaw 2026.3.13 lets your agent watch along in your real browser via Chrome DevTools MCP.

Edition #6. Anthropic makes 1 million context tokens available to everyone. OpenClaw 2026.3.13 lets your agent watch along in your browser. Nvidia is working on NemoClaw, an open-source AI agent platform for enterprises. Google Maps now just answers your questions.

Claude Opus and Sonnet get 1 million tokens of context

Anthropic is making the full 1 million tokens of context available for Claude Opus 4.6 and Sonnet 4.6. No more beta header needed, no surcharge for long context, standard pricing across the entire window. Opus 4.6 scores 78.3 percent on MRCR v2 at 1 million tokens, the highest of any frontier model.

What that means in practice: you can load a complete codebase, thousands of pages of contracts, or the full trace of a long-running agent in one go. No more lossy summaries, no context-clearing. Media limits go from 100 to 600 images or PDF pages per request.

For Claude Code users with Max, Team or Enterprise: Opus 4.6 sessions automatically use the full 1M window. Fewer compactions, more of the conversation stays intact.

OpenClaw 2026.3.13: your agent watches along in your browser

OpenClaw has an update worth mentioning. Version 2026.3.13 introduces live Chrome session attach. Your agent can now watch along in your real browser, with your logins, cookies and open tabs. No extensions needed.

How it works: open chrome inspect with remote-debugging in Chrome, tick it on, done. Your agent uses Chrome DevTools MCP under the hood and sees everything you see. Think of: an agent that reads your inbox, fills in forms or does research in your logged-in sessions.

Also in this update: Android app fully redesigned and brought back to 7MB, iOS gets a welcome pager, Docker timezone override, and Windows gateway improvements.

Nvidia is working on NemoClaw: open-source AI agents for enterprises

Nvidia is working on NemoClaw, an open-source platform with which enterprises can deploy AI agents that independently execute tasks for their employees. Plan, reason, handle multi-step workflows. According to Wired it has already been pitched to Salesforce, Cisco, Google, Adobe and CrowdStrike.

Nvidia has been selling the chips on which AI runs for years. Now they want the software too. NemoClaw will be open source, contains security and privacy tools and works independently of your hardware. It's Nvidia's enterprise answer to OpenClaw, which runs locally for individual developers. NemoClaw aims at companies that want to roll out agents across their entire organization, with compliance and security built in.

Jensen Huang recently called OpenClaw the most important software release probably ever. With NemoClaw Nvidia wants to own that same thing for the enterprise market.

Google Maps now answers your questions

Google is coming out with Ask Maps, an AI layer on top of Google Maps. You ask a question in natural language, Maps searches 300 million locations and 500 million reviews, and gives a personalized answer. From that answer you can directly book or navigate. No more list of search results, but a conversation that understands what you mean. US and India first, the rest follows.

How I made AI work for me this week

The power isn't in the tool itself, but in what you build around it. I run OpenClaw on a dedicated European server and have built software around it that's specifically made for my agents. Scrapers, a sales system, CRM connections, bookkeeping. That's where you make the difference if you use OpenClaw.

Tarik Eraslan

Written by

Tarik Eraslan

Founder of AI Agent. Helps businesses implement AI in their daily workflows.

LinkedInMore about Tarik

Ready to deploy AI?

Schedule a call to discuss your AI Agent or explore our Academy.

Claude's 1 million tokens of context is live. And your…