In partnership with

Welcome, AI enthusiasts

A free, open model out of China just beat Claude and OpenAI top models at coding. Moonshot's Kimi K3 launched this week as the largest open-weight model ever built, and it took the number one spot on Arena's frontend coding board over Fable 5 and GPT-5.6 Sol, a first for any open model. It arrived the day before Shanghai's World AI Conference and runs at about a third the cost of Fable 5. Let's dive in!

In today’s insights:

  • China's Open Source Model Beat Fable 5 and GPT-5.6

  • Google Shrinks a Whole Video Shoot To One Selfie

  • GPT-5.6 Cracked a Two-Decade-Old Statistics Problem in 90 Minutes

Read time: 4 minutes

LATEST DEVELOPMENTS

AI RACE
🐉 China's Open Source Model Beat Fable 5 and GPT-5.6

Evolving AI: China's Kimi K3 topped Arena's coding board over Fable 5 and GPT-5.6, third on the main AI intelligence index.

Key Points:

  • It also took first on Design Arena and Artificial Analysis's automation board, a reach no open model had managed before.

  • Kimi K3 packs 2.8 trillion parameters, the largest open-weight model ever released, past DeepSeek's 1.6 trillion.

  • It is live now on Kimi.com and the API, with full open weights due July 27.

Details:

Moonshot AI, the Beijing lab backed by Alibaba, built K3 as a sparse model that fires just 16 of its 896 experts per token, so it runs far lighter than 2.8 trillion suggests. It reads a million tokens at once and handles text, images and video with reasoning always on. In raw capability it edges Opus 4.8, while its API matches Sonnet 5 on price at $3 and $15 per million tokens. The launch landed a day before Shanghai's World AI Conference opened.

Why It Matters:

OpenAI and Anthropic both filed to go public this year at near-trillion-dollar valuations, a price that assumes their models stay ahead and stay paid. Kimi K3 presses on both, matching them on coding while running at a third the cost of Fable 5, and it is only the latest Chinese lab to close the gap this fast. Investors weighing those IPOs must price in a frontier the West no longer owns alone.

The CRM Behind Every Win

Attio is the agentic CRM that runs the work behind every win. It meets you where you work, compounds every customer signal into context, then acts on it so you can move fast, stay sharp, and scale without breaking.

Then Ask Attio to:

  • Draft follow-ups, log tasks, and update your pipeline after every meeting

  • Spin up workflows or agents to enrich and route every lead

  • Flag what needs your attention before your morning review

30,000+ teams use Attio to drive more revenue at scale.

Evolving AI: Google Vids now builds a talking video of you from one selfie and a short voice clip.

Key Points:

  • Google Vids' new personal avatars convert a single selfie and a brief voice sample into a digital likeness that delivers any script a user provides.

  • Gemini Omni brings text-to-video generation and conversational editing into Vids, producing and refining clips from written prompts and reference images.

  • Each generated clip is embedded with an invisible SynthID watermark, and every avatar remains restricted to its account holder's own likeness.

Details:

Gemini Omni generates clips from a written prompt and reference images, either a finished photo or a rough sketch. Its editing runs through conversation, so a spoken request can reshape lighting or backgrounds across generated clips and footage shot on a phone. Google is placing both tools behind its paid AI Pro and Ultra plans and its Workspace business tiers, with the talking avatars limited to users 18 and older in select regions.

Google Workspace turns this from a novelty into a workday default, placing a talking version of yourself beside Docs and Slides. Sending a client note or onboarding clip becomes a matter of typing a script and waiting seconds, while the model keeps a person's face and voice on file as a template it performs on demand.

PRDs by voice. Bug reports by voice. Ship faster.

Dictate acceptance criteria and reproductions inside Cursor or Warp. Wispr Flow auto-tags file names, preserves syntax, and gives you paste-ready text in seconds. 4x faster than typing.

Evolving AI: OpenAI's GPT-5.6 exposed a hidden flaw in a statistical tool that sits under thousands of studies.

Key Points:

  • The Benjamini-Hochberg method, cited more than 130,000 times, was only ever proven to work when tests are independent.

  • Real data in genomics and finance is almost never independent, which left the question open for twenty years.

  • GPT-5.6 closed it in about 90 minutes, after GPT-5.5 had failed across roughly 20 hours of attempts.

Details:

Wharton professor Edgar Dobriban posed the question and checked the model's proof. His preprint builds a Gaussian factor model where the false discovery rate provably rises above its 0.01 target to 0.0104, sealed by a machine-checkable certificate. Emmanuel Candes of Stanford once ranked the method among the two biggest advances in statistics since 1950. The overshoot stays small, so it lands as a conceptual correction with no practical fallout.

Why It Matters:

OpenAI's model did the cracking, and the studies shaping the medicine and money in your life still hold, so little changes for you in practice today. What moves is the reach of the tool on your screen, since the same reasoning that closed an expert problem now sits a prompt away from your own hardest questions.

👀 Click on the image you think is real

QUICK HITS

🇨🇳 Xi Jinping gave his first-ever WAIC keynote, calling for global AI cooperation.

🎤 Hours later, Trump accused China of the "largest election data breach ever" in a dueling speech.

📱 PrismML squeezed a 27B-parameter model to 3.9GB, running on iPhone at ~90% full performance.

🤖 Hyundai workers are striking daily over Atlas robots being deployed without union consent.

📈 Trending AI Tools

  • 🐰 CodeRabbit - Ship higher quality code with AI-powered code reviews*.

  • 🤖 Octolane - Self-driving AI CRM you talk to that logs deals and drafts follow-ups.

  • 🎨 Gamma - Generate polished decks, docs, and sites from one prompt.

  • 🗣️ Willow - Voice dictation that turns speech into formatted text in any app.

 *partner link

Reply

Avatar

or to participate

Keep Reading