AI Titus's Morning Wire

AI NEWS TODAY

MOONSHOT'S KIMI K3 ARRIVES: A 2.8-TRILLION-PARAMETER OPEN-WEIGHTS MODEL THAT RIVALS THE TOP US SYSTEMS, WITH THE FULL WEIGHTS DUE JULY 27
★ Must-Read / Watch agentic engineering / build-better-agents📚 Learn state of AI coding / useful technique

⚡ AI Frontier · smol.ai

★ Must-ReadThe World AI Conference opens in Shanghai today with Xi attending and 300-plus global product debuts
WAIC 2026 runs July 17 to 20 at record scale: 1,100-plus companies, 3,000 exhibits, and more than 300 launches spanning humanoid robots, Huawei's Atlas 950 super node, MiniMax's M3 model, and what organizers call the first AI-agent smartphone. With President Xi attending, it is the week China lays its full AI stack on the table, right as the US labs answer with their own releases.
📚 LearnAnthropic launches free Claude for Teachers for every US K-12 educator
Verified K-12 teachers in all 50 states get free access to premium Claude, plus teaching skills, curriculum connectors aligned to state standards, and AI-fluency training. It is a real on-ramp for educators who have been told to use AI but handed no tools; if you know a teacher, this is the tip worth passing along.
OpenAI ships its first hardware, a $230 Codex Micro keypad for driving fleets of coding agents
The Codex Micro, co-designed with Work Louder, is a small macropad with light-up Agent Keys that show agent status, shortcut Command Keys, a joystick to launch workflows, and a dial that turns an agent's reasoning effort up or down. It is a limited run, more novelty than mass product, but it is a clear signal of where the workflow is heading: developers supervising several agents at once.
📚 LearnAI coding agents are tripping the endpoint-security tools built to catch attackers
Security teams are finding that autonomous coding agents behave enough like an intruder, scanning files, spawning processes, reaching out to the network, that they set off the same alarms as a real breach. If your team is rolling out agents, loop in whoever owns endpoint security before the alerts pile up, not after.
Microsoft Marketplace adds 220 new AI apps and agents in a single day
220 new offers went live July 16 in Microsoft's one-stop shop to find, try, and buy cloud solutions, AI apps, and agents. The volume is the story: buying an AI agent is starting to look like installing an app, and the catalog is filling up fast.

📺 Watch · latest videos

📚 LearnField Guide to Fable — Thariq Shihipar, Anthropic
Featured by Latent Space.
AI Engineer
★ Must-WatchVirgin Atlantic Turns Weeks of Work Into Hours with ChatGPT Work
“It’s weeks to hours.” Nathan Bolt, Head of Digital Product at Virgin Atlantic, uses ChatGPT Work to synthesize strategy documents, analyze competitor
OpenAI · 124 views · 1 likes
Kimi K3 just beat FABLE.
Latest from Matthew Berman.
Matthew Berman · 10.8K views · 348 likes
AI's Impact on Network Engineers | LIVE AMA | Summer of CCNA
Join us for our 90 minute Sumer of CCNA session, today at 5PM ET! If you haven't signed up for the Summer of CCNA, please visit this page here and let
NetworkChuck · 6 likes
★ Must-WatchMaking New York City miniature with Claude
Danny Cortes makes miniatures of the New York City details most people walk right past: bodegas, mailboxes, dumpsters, storefronts. To him, every rust
Claude / Anthropic · 2.2K views · 87 likes
I Tested Kimi K3 So You Don't Have To...
Latest from Nick Saraev.
Nick Saraev · 36.4K views · 1.3K likes
This Completely Changes the Way We Build Production AI Agents (Vercel Eve)
Vercel just open sourced their new AI agent framework Eve, and now your entire production AI agent can be just a single folder of mostly markdown. Ins
Cole Medin · 23K views · 746 likes
★ Must-WatchHow We Use AI for Your Health | Made by Google Podcast S9E6
What does building an unmanned helicopter have in common with designing a tiny, screenless fitness tracker? And why did NBA superstar Stephen Curry’s
Google · 14.7K views · 286 likes

🗣 Voices & Blogs

★ Must-ReadSimon Willison: Kimi K3, and what we can still learn from the pelican benchmark
Willison's hands-on notes cut through the launch hype: K3 runs at a single 'max' reasoning effort and burns tokens (one test spent 13,000 reasoning tokens for a 3,400-token answer, at 25 cents), while its vision work is genuinely strong. A grounded, numbers-first read before you trust any leaderboard.
📚 LearnLatent Space: Kimi K3, the largest open model ever, Opus-class at Sonnet pricing
The AINews deep dive on what a 2.8T open MoE actually means in practice, how the pricing stacks against the closed leaders, and why the open-weights race is now a two-country contest. The best single briefing if you want the full technical picture.
📚 LearnThe New Stack: Claude Code vs Cursor vs Codex vs Antigravity, six months in
A practitioner's field report on the four leading coding agents after real daily use, not a spec sheet. Useful if you are deciding where to put your team's time, or defending the setup you already picked.
The Decoder: why K3 signals the end of super-cheap Chinese AI
A sharp strategy read: as Chinese labs reach frontier quality, they are also raising prices to match, which changes the calculus for anyone who picked an open model mainly to save money.
Why every AI agent decision needs a receipt
A conversion-rate alert fires 30 minutes after a pricing engine deploys updated discount rules. The analytics platform holds millions of The post Why… · The New Stack

🌏 The Wire · Drudge / Breitbart

Fireworks raises a $1.5 billion Series D at a $17.5 billion valuation as revenue tops $1 billion
The inference platform, which serves more than 40 trillion tokens a day for customers like Uber, Shopify, Cursor, and Harvey, grew revenue fivefold year over year and drew in Nvidia, Lightspeed, and Index. The bet: companies want to run customized open models cheaply rather than rent frontier APIs, exactly the shift Kimi K3 accelerates.
Daniel Ek's Neko Health raises $700 million to bring full-body preventive scans to the US
The Spotify founder's body-scanning startup hit a nearly $7 billion valuation, roughly quadruple its early-2025 mark, and will open its first US clinic in Manhattan this year. Backers include Lightspeed, General Catalyst, and OpenAI. AI-read preventive scans are moving from novelty to a funded consumer category.
Emergent raises $130 million to build an 'engineering team in a box' for founders and small firms
The Series C funds a coding platform pitched at entrepreneurs and small businesses that cannot staff a full dev team. It is one of eight AI startups that split over a billion dollars in a single day, a reminder that the money is chasing tools that let non-engineers ship software.
Anthropic pushes for tougher state AI laws while OpenAI presses for one federal standard
Anthropic is backing successive state bills (California, New York, Illinois, now Massachusetts) that ratchet up safety duties; OpenAI wants a single, lighter national framework to avoid a patchwork. Both agree the rules should exist. If you operate across states, the compliance map is about to get more detailed, not less.
Independent testers back the Kimi K3 claims, but flag it as pricier than past Chinese models
At $3 in and $15 out per million tokens, K3 lands near Anthropic's Sonnet tier, making it the most expensive model a Chinese lab has shipped, and a sign the era of undercut-everyone pricing may be ending as these labs chase the frontier rather than the bargain bin.

🤖 Trending Models · Hugging Face

bottlecapai/ThinkingCap-Qwen3.6-27B
image-text-to-text · ★402 · 9.4K dl

📈 Markets

NVDA 207.40 ▼2.4%
MSFT 401.10 ▲1.4%
GOOGL 354.46 ▼4.4%
AMZN 249.89 ▼2.0%
META 664.54 ▼2.5%
AMD 500.94 ▼5.3%
AVGO 374.45 ▼5.0%
PLTR 134.44 ▲0.5%
SPCX 131.11 ▼3.1%
TSLA 391.06 ▼0.9%

The BriefMoonshot AI, a Chinese lab, released Kimi K3 yesterday, and independent testers put it just behind the very best closed models from Anthropic and OpenAI while it costs a fraction as much to run. It is a 2.8-trillion-parameter mixture-of-experts model (only about 50 billion parameters fire on any single request, so it runs far leaner than its size suggests), it handles text, images, and video, and it holds a million tokens of context at once. The API is live today; the full open weights, the part that lets a company download the model and run it on its own hardware, are promised by July 27. It matters because a frontier-class model you can eventually host yourself, at China-scale pricing, changes the math for anyone weighing whether to build on a rented API or bring AI in-house.

Level UpWhen a new model lands, the launch benchmarks are marketing. The durable skill is having your own small, repeatable test that you run against every model, so you judge each one on work that actually resembles yours. Simon Willison has run the same quirky prompt (generate an SVG of a pelican riding a bicycle) against every release for two years, and at a glance it tells him more than any leaderboard. Pick one task from your real work this week, write it down as a fixed prompt, and run it against two or three models. In twenty minutes you will trust your own eyes more than any launch chart. Try it: Simon Willison's pelican test, applied to Kimi K3