Moonshot AI reveals a frontier-level model

Welcome back, Superhuman. Just when you think the AI race can’t get any more competitive, yet another lab elbows its way in. This time, it’s Moonshot AI’s turn. The Chinese lab just revealed Kimi 3 — a cost-efficient model that tops Anthropic’s Fable 5 on design (and a few other benchmarks).

Today: 29 countries form new AI org, how to create a daily work brief using ChatGPT Work, and get the latest prompts and trending social posts.

1. Moonshot AI reveals Kimi K3, a major jump in Chinese frontier performance: Kimi K3 is the lab’s most capable model yet, offering cost-efficient frontier intelligence for coding, knowledge work, and general reasoning — notably topping Anthropic’s Fable 5 and OpenAI’s GPT-5.6 to claim the #1 spot on Arena’s leaderboard for web design. The Chinese lab is reportedly raising fresh capital in a round that would value it at $31.5B. Watch the viral launch video (1M views).

2. Twenty-nine countries sign an AI cooperation organization: The World AI Cooperation Organization is a new intergovernmental body headquartered in Shanghai. Founding members include China, Russia, Serbia, Belarus, Cuba, Brazil, and Venezuela, alongside 10 African and 12 Asian nations — notably none from the US or Western Europe.

3. Canva launches a major upgrade to its AI design tool: Canva Code 2.0 lets you vibe-code websites and apps from a plain-language prompt — then edit the output like any other Canva design. The platform also lets you launch a design from any starting point, including pre-built templates, HTML, or other design projects, blurring the line between graphic design and coding. Now available to all Canva members, including free plans. See how it works.

REST serves the developers building against your API. MCP serves the agents now trying to use it. Most teams treat these as competing standards and have to pick one.

They're not rivals, they're layers: most MCP servers just call REST internally to do the real work. The best ones don't convert every endpoint into a tool, they design around what the agent's trying to accomplish.

Shipping that server also means shipping OAuth 2.1 with scoped tokens. WorkOS AuthKit already speaks that spec, so you skip building an auth provider on top.

Read the breakdown →

Apple sued OpenAI in federal court over alleged trade secret theft. The suit alleges that OpenAI is running a coordinated poaching campaign through two former Apple employees now on its payroll: Chang Liu and Tang Tan. Apple claims the AI lab is targeting its employees and coaching them to steal secrets, calling its hardware division “rotten to its core.” OpenAI denies all of it.

Google DeepMind’s CEO proposed a regulatory blueprint for AI. Demis Hassabis — one of AI’s leading voices — published a rare thought piece this week, calling for the US to create a self-regulatory Standards Body to track progress and review new AI models. It would function similarly to FINRA, the private watchdog that sits between Wall Street and the SEC. Read his full essay.

Thinking Machines put America on the open-weight board. Mira Murati's startup shipped Inkling, its debut model that’s both powerful, efficient, and open-weight — meaning companies can download and customize it. That’s a major strategy shift for a US lab, with most open-weight models so far coming from China. The timing is favorable, with many American companies pushing to own more of their AI stack.

AI executives say demand is still outpacing supply. Former Intel CEO Pat Gelsinger told CNBC that demand for AI is still "almost unlimited.” Nebius CRO Marc Boroditsky agreed, saying there’s "much more demand than we're able to fill." But both flagged a separate trend: companies are getting more cost-conscious about their AI spending.

Sample Prompt: Build my daily work brief using [@Google Calendar], [@Gmail], [@Slack], and [@Google Drive]. Review today’s meetings, unread messages, follow-ups, and project notes. Summarize my top priorities, meeting prep, replies needed, decisions, important FYIs, and missing context. Do not send messages, edit files, or create tasks.

Recurring Prompt: Prepare my daily work brief every weekday at [time]. Check the same sources and only highlight new or changed priorities, meetings, replies, decisions, and blockers. Do not send messages, edit files, or create tasks.

Quick Tip: Pin the task for quick access

Using AI a lot isn't the same as being good at it. Those who are fluent reclaim 30% of their time while raising quality. Watch Clarinet's on-demand event on the skills and mindset behind real AI fluency, and how to measure where your team stands.

Watch now.

📝 Prompting Tip: Anthropic engineer Thariq Shihipar shared his ideal prompting strategy, which uses thin prompts with meaty context. It’s sparked a discussion on prompting in the replies.

✈ Afterburn: One Redditor called this vibe coded fighter jet game “the most ridiculous and insane thing I've seen all day.” It rivals Microsoft Flight Simulator, but was built entirely with Claude Code (2K upvotes).

👀 Kimi in Action: One AI power user already tested Kimi K3 against Opus 4.8 in a one-prompt challenge. The task? Create a scene from a video game. The comments agree that Kimi K3 won — and it wasn’t very close.

🤦 AI Deepfake: Did you see that viral video of Norwegian soccer player Erling Haaland getting scared of his reflection? It was likely AI-generated, according to a follow-up post. See the side-by-side of both videos (13M views).

🟰 Solving Problems: One of the world’s leading theoretical physicists says Claude Fable solved a problem that he was stuck on for the past 6 months (2.7K upvotes).

^{* indicates a promoted tool, if any}

OpenAI shipped Codex Micro, a physical keypad that swaps the full keyboard for a handful of one-command buttons that make it easier to command agents.

Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.

Until next time — Zain, Theodore, & the Superhuman AI team.