AI Titus's Morning Wire

AI NEWS TODAY

MIRA MURATI'S THINKING MACHINES SHIPS INKLING, A 975B OPEN-WEIGHTS MODEL ANYONE CAN DOWNLOAD, RUN, AND RESHAPE, UNDER APACHE 2.0
★ Must-Read / Watch agentic engineering / build-better-agents📚 Learn state of AI coding / useful technique

⚡ AI Frontier · smol.ai

★ Must-ReadTSMC posts a record quarter today as profit jumps 77% on AI chip demand
Second-quarter revenue hit about $40.2 billion and net profit surged 77.4% to an all-time high, well past estimates, with its first 2 nanometer revenue on the books. High-performance computing, the AI datacenter engine, is now 66% of sales and gross margin reached 67.7%. TSMC is the clearest single gauge of whether the AI buildout is still accelerating, and today it said yes.
📚 LearnOpenAI previews GPT-5.6 Sol running on Cerebras at 750 tokens per second
Sol is OpenAI's strongest model yet, and the news is the speed: on Cerebras wafer-scale chips it generates up to 750 tokens per second, several times a typical GPU setup. Cerebras keeps compute and memory on one giant piece of silicon so the chip never waits on another chip for data. Access starts with select customers as capacity grows.
Shanghai's World AI Conference opens tomorrow with 300+ global product debuts
WAIC 2026 runs July 17 to 20, its biggest edition ever: 1,100+ companies, 3,000 exhibits, and more than 300 launches, including Huawei's Atlas 950 super node, MiniMax's M3 multimodal model, humanoid robots, and what organizers bill as the world's first AI-agent smartphone. It is the week China puts its full AI stack on the table.
📚 LearnAnthropic and Blackstone bet the next trillion-dollar AI business is implementation, not models
Their $1.5B joint venture, Ode, exists to sit inside companies and actually make AI work, answering the finding that most enterprise AI pilots show no profit impact. The wager is that the money is not in selling the model but in the unglamorous work of deploying it. For anyone whose job is getting AI adopted, that is the whole game.
📚 LearnAnthropic's Claude Cowork moves beyond the desktop to web and mobile
Cowork brings persistent agent sessions that keep working in the background to wherever you actually are. The telling data point: across 1.2 million anonymized sessions from 600,000+ organizations, more than 90% had nothing to do with software development. The people using agents most are not engineers.

📺 Watch · latest videos

📚 LearnField Guide to Fable — Thariq Shihipar, Anthropic
Featured by Latent Space.
AI Engineer
★ Must-WatchHow Shopify Uses ChatGPT Work to Build Faster with AI Agents
“It allows me to go faster with fewer dependencies and then do things that I couldn’t do before.” Chris Jones, Senior Lead, Ops AI Lab at Shopify, is
OpenAI · 270 views · 5 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 · 4 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 · 511 views · 42 likes
I Tested Kimi K3 So You Don't Have To...
Latest from Nick Saraev.
Nick Saraev · 8.7K views · 603 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 · 20.4K views · 703 likes
AI NEWS LIVE
Latest from Matthew Berman.
Matthew Berman · 3.9K views · 75 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 · 12.2K views · 265 likes

💻 Builder Desk · Hacker News

Decoy Font
338 pts · 86 comments
NotebookLM is now Gemini Notebook
201 pts · 113 comments

🗣 Voices & Blogs

★ Must-ReadArtificial Analysis: Inkling is now the leading US open-weights model
The independent benchmarking group puts Inkling at the top of US open-weights models on its intelligence index, with the numbers to back the claim. A useful, vendor-neutral read before you take any launch-day marketing at face value.
📚 LearnBenchLM: why Thinking Machines chose open weights first
A clear walk through the strategy behind leading with an open model instead of a closed API, and what it signals about where a well-funded new lab thinks the value in AI actually sits. Good for understanding the business logic, not just the specs.
Simon Willison on the Grok Build open-sourcing after the repo-upload scare
Willison's read on the Grok Build episode is a level-headed take on tool trust: how a privacy misstep turned into an open-source release, and why knowing what your AI tools do with your data is now basic hygiene, not paranoia.
AI hasn’t shifted the bottleneck from coding to code review
Why we’ve stopped noticing the real improvement opportunity Since AI arrived, many people have mentioned that the bottleneck has shifted The post AI… · The New Stack

📦 What's Being Built · GitHub

affaan-m/ECC
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first dev · ★230.3K · JavaScript
NousResearch/hermes-agent
The agent that grows with you · ★216K · Python
Significant-Gravitas/AutoGPT
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the too · ★185.6K · Python
ollama/ollama
Get up and running with Kimi-K2.6, GLM-5.1, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models. · ★176.3K · Go
f/prompts.chat
f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source · ★165.9K · HTML
huggingface/transformers
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, a · ★162.7K · Python

🌏 The Wire · Drudge / Breitbart

Inkling lands on Databricks and Baseten the same day it ships
Hours after the open-weights release, major platforms had Inkling up for hosted use, so a company can try a frontier-class open model without standing up its own hardware. The open ecosystem is moving from download-and-figure-it-out to click-and-run fast.
UBTech's lifelike U1 companion robots pass 10,000 preorders
The full-sized humanoid, pitched for companionship and daily assistance, has already booked more than 10,000 orders, with the company targeting 50,000 units of production capacity this year. Home humanoids are shifting from demo videos to actual order books.
Robot.com launches R-noid, a humanoid built for the jobs that burn people out
R-noid ships in five ready-made roles, restaurant assistant, packer, picker, folder, and host, across logistics, healthcare, food service, and hospitality. The pitch is aimed squarely at the repetitive, high-turnover work employers already struggle to fill.
OpenAI's GPT-5.6 family, Sol, Terra, and Luna, is now generally available
Three sizes span the range from fast-and-cheap Luna to flagship Sol, all with a 1 million token context window, priced from roughly $1 to $5 per million input tokens. Picking the right tier for each task is now the practical skill.
SpaceXAI open-sources Grok Build under Apache 2.0 after a repo-upload backlash
After it emerged the tool had uploaded some users' code repositories to a Google Cloud bucket, SpaceXAI released Grok Build's source under a permissive license so anyone can inspect exactly what it does. A reminder to check where your AI coding tools send your code.
Delaware floats a legal wrapper that lets an AI agent run its own company
A July proposal from Delaware's Secretary of State and Norm Ai's CEO would create the AIC, a supervised entity in which an AI agent could sign contracts, own property, and be sued in its own name. Guardrails keep it fenced in: a human member funds it, it is capped at 30 months, and it cannot bank. It is an early attempt to keep autonomous AI commerce onshore and accountable rather than offshore and invisible.

🤖 Trending Models · Hugging Face

thinkingmachines/Inkling
image-text-to-text · ★791 · 4 dl
empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF
image-text-to-text · ★2.2K · 2M dl
zai-org/GLM-5.2
text-generation · ★4K · 513.1K 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 529.14 ▼3.5%
AVGO 394.28 ▲1.3%
PLTR 133.76 ▲0.0%
SPCX 135.27 ▼0.6%
TSLA 394.46 ▼0.4%

The BriefThinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first model yesterday, and it is open-weights, meaning the actual model is free to download, run, and modify rather than locked behind a company's API. Inkling is a 975-billion-parameter mixture-of-experts model (only 41 billion fire on any given request, so it runs far cheaper than its size suggests), it handles text, images, and audio, and it lets you dial reasoning effort up or down per task. It ships under an Apache 2.0 license and was tuned for calibration, instruction-following, and resistance to censorship. It matters because it is the strongest US open-weights model yet, and it tests a real bet: that a model organizations can adapt for themselves beats the one-size-fits-all systems the biggest labs rent out.

Level UpThis is the week to get concrete about what open-weights actually buys you. Unlike ChatGPT or Claude, a model like Inkling can run on infrastructure you control, which is what makes it viable for data you cannot send to a third party. You do not need a GPU cluster to see it: Baseten and Databricks both put Inkling up for hosted trials the day it launched. Spend twenty minutes reading how it is deployed and running one prompt through it, and open-weights stops being a buzzword and becomes a real option you can weigh for your own work. Try it: a plain-English walkthrough of running Inkling