Titus's Morning Wire

AI Titus News

WEDNESDAY, JULY 1, 2026
US LIFTS EXPORT FREEZE ON ANTHROPIC'S TOP MODELS: FABLE 5 AND MYTHOS 5 RETURN WORLDWIDE JULY 1 AS SONNET 5 SHIPS CHEAP
★ Must-Read / Watch agentic engineering / build-better-agents📚 Learn state of AI coding / useful technique

⚡ AI Frontier · smol.ai

★ Must-ReadAnthropic launches Claude Sonnet 5 as the new default, near-Opus quality at a fraction of the cost
Anthropic released Claude Sonnet 5 on June 30 as the new default for Free and Pro users, calling it its most agentic mid-tier model yet with performance approaching Opus 4.8 at far lower cost. Intro pricing is $2 per million input tokens and $10 per million output through August 31 (then $3 and $15), live across Claude, Claude Code, the API, and Amazon Bedrock.
★ Must-ReadAnthropic unveils Claude Science, an AI workbench for reproducible research
Announced June 30 at Anthropic's 'AI for Science' briefing, Claude Science is a workbench that plugs into researchers' own tools and packages and produces auditable, rerunnable results with flexible compute access. It targets the reproducibility problem in computational research rather than one-off chat answers.
★ Must-ReadOrnith-1.0: open-weights, self-scaffolding models built for agentic coding
A new MIT-licensed, open-weights model family ranging from about 9B to 397B parameters, designed for self-scaffolding agentic coding where the model plans and structures its own multi-step work. Open weights let teams run it locally or fine-tune it, a sharp contrast to this month's locked-down frontier previews.
📚 LearnGoogle ships Nano Banana 2 Lite: four-second images at three cents each
Google launched Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) on June 30 via the Gemini API, AI Studio, and Gemini Enterprise. It generates images in about four seconds at roughly $0.034 each, with strong character consistency and legible text rendered inside the image.
SpaceX signs a compute lease reported at $6.3B with open-source startup Reflection AI
SpaceX signed a multi-year compute lease reported at $6.3 billion with open-source startup Reflection AI, with payments beginning July 1 at about $150 million per month. Reflection will train frontier open-weight models on Nvidia GB300 capacity at SpaceX's Colossus 2 data center near Memphis.
Two bets on releasing frontier AI: OpenAI's GPT-5.6 stays government-gated as Sonnet 5 goes global
As Anthropic pushed Sonnet 5 out globally, OpenAI's GPT-5.6 family stayed restricted to roughly 20 government-approved partners at US request. The split highlights two very different strategies for shipping the most capable models: open and cheap versus gated and vetted.

📺 Watch · latest videos

★ Must-WatchClaude Sonnet 5 Just Dropped (I have to be honest...)
An honest hands-on review of Claude Sonnet 5, weighing where the cheaper mid-tier model actually holds up against the flagship and where it does not.
AI Foundations
★ Must-WatchAI News: An INSANE Week, Here's What Matters
A fast roundup of a huge week in AI: Sonnet 5, the Anthropic export-control reversal, Claude Science, and the GPT-5.6 gating story, and what each means for builders.
AI News
📚 LearnClaude Sonnet 5 + Claude Code Is Unfair (INSANE)
A working demo of Sonnet 5 inside Claude Code, running multi-step coding tasks to show what the new default model can do in a real agentic workflow.
AI Builder
📚 LearnClaude Sonnet 5 just dropped. I'm changing how I use AI
A walkthrough of the practical changes Sonnet 5 brings and how the creator is rethinking which model to reach for now that near-flagship quality is cheap.
AI Review

💻 Builder Desk · Hacker News

Claude Science
534 pts · 153 comments
Godot will no longer accept AI-authored code contributions
439 pts · 283 comments
Nano Banana 2 Lite
417 pts · 172 comments
Asahi Linux 7.1 Progress Report
373 pts · 105 comments
Why I Stopped Arguing with People
319 pts · 240 comments

🗣 Voices & Blogs

★ Must-ReadSimon Willison: Anthropic export controls lifted on Claude Fable 5 and Mythos 5
Willison's same-day writeup of the Commerce Department lifting export controls on Fable 5 and Mythos 5, set against his earlier coverage of the June 12 suspension. A clear-eyed read on how AI safety and export policy are now colliding.
★ Must-ReadSimon Willison: What's new in Claude Sonnet 5
A practical first look at Sonnet 5, including the detail that its new tokenizer produces roughly 30 percent more tokens than Sonnet 4.6, which matters if you budget by token. Worth reading before you switch workloads over.
📚 LearnSimon Willison: Ornith-1.0, self-scaffolding LLMs for agentic coding
Willison digs into Ornith-1.0, the new open-weights family built for agentic coding, and what self-scaffolding models mean for people who run their own tooling. A grounded take on why open weights still matter.
Jon Udell: it hurts when agents create unreviewable PRs
Jon Udell on a growing problem: coding agents that produce pull requests too large or opaque for a human to meaningfully review. A useful prompt to keep humans genuinely in the loop, not just nominally.
The call is coming from inside your pipeline: the anatomy of a Codecov attack
In January 2021, an attacker added a single line of code to a popular bash script. Tens of thousands of The post The call is coming from inside your… · The New Stack
CyberGym-E2E: Scalable Real-World Benchmark for AI Agents' End-to-End Cybersecurity Capabilities
Agentic AI Summit Agenda Now Live + New Startup Spotlight Session (Apply Today!) | CyberGym-E2E Research Highlight: Can AI Agents Run the Full… · via Berkeley RDI · Agentic AI Weekly

📦 What's Being Built · GitHub

affaan-m/ECC
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first dev · ★224.5K · JavaScript
NousResearch/hermes-agent
The agent that grows with you · ★207K · 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 · ★185.2K · Python
ollama/ollama
Get up and running with Kimi-K2.6, GLM-5.1, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models. · ★175.2K · Go
f/prompts.chat
f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source · ★164.6K · HTML
huggingface/transformers
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, a · ★162.1K · Python

🌏 The Wire · Drudge / Breitbart

US lifts Anthropic export freeze, and a new industry jailbreak-severity framework arrives
The US Commerce Department lifted its 18-day export freeze on Anthropic's Fable 5 and Mythos 5, imposed after Amazon researchers found a way to bypass the models' safety controls. Anthropic says a single new safety filter now blocks the risky prompts, and it is publishing an industry jailbreak-severity scoring framework with Amazon, Microsoft, and Google.
NASA readies LINK, a first-of-its-kind mission to boost the aging Swift observatory's orbit
NASA and Katalyst Space are launching LINK, a commercial servicer that will dock with the 21-year-old Swift observatory and raise its orbit to extend its science life. The July 1 launch from Kwajalein Atoll on a Pegasus XL rocket follows a June 30 weather scrub.
New catalyst turns industrial waste heat into clean hydrogen at far lower temperatures
University of Birmingham researchers built a perovskite catalyst that converts industrial waste heat into hydrogen fuel at temperatures well below the 1300 to 1500 C that today's systems need to regenerate. That could let steel mills, cement works, and renewable sites make clean hydrogen from heat they already throw away.
Astronomers confirm two 'super-puff' planets less dense than cotton candy
Astronomers confirmed two 'super-puff' planets roughly the size of Jupiter but less dense than cotton candy, among the puffiest worlds ever found. Their bloated, low-gravity atmospheres make them ideal targets for telescopes studying how planets form.
Einstein Probe may have caught a black hole tearing apart a white dwarf star
The Einstein Probe X-ray telescope may have observed a black hole shredding a white dwarf, one of the rarest and most dramatic events in the cosmos. If confirmed, it gives astronomers a rare live look at the extreme physics around black holes.
Ancient collision may let future astronauts walk on rocks from deep inside the Moon
A new study finds an ancient impact may have exposed rocks from deep inside the Moon near future landing sites, meaning astronauts could sample material from the Moon's hidden interior on foot. It would open a direct window into how the Moon formed without having to dig.
AI-assisted map finds 5,800+ square miles of climate-resilient 'super reef' refugia
An AI-assisted analysis of 65 years of coral data identified more than 5,800 square miles of climate-resilient 'super reef' refugia across 71 countries, roughly triple the resilient area found in the landmark 2018 study. The map gives conservation groups a shortlist of the reefs most worth protecting.

🤖 Trending Models · Hugging Face

baidu/Unlimited-OCR
image-text-to-text · ★1.6K · 630.2K dl
empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF
image-text-to-text · ★1.1K · 1.1M dl
zai-org/GLM-5.2
text-generation · ★3.1K · 160K dl
Qwen/Qwen-AgentWorld-35B-A3B
text-generation · ★486 · 34.4K dl
yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF
text-generation · ★908 · 288.7K dl

📈 Markets

NVDA 200.09 ▲2.6%
MSFT 373.02 ▲1.2%
GOOGL 357.37 ▲1.1%
AMZN 238.34 ▼0.7%
META 563.29 ▲0.1%
AMD 580.91 ▲7.7%
AVGO 377.75 ▲1.4%
PLTR 116.67 ▲0.8%
SPCX 170.86 ▲4.1%
TSLA 420.60 ▲2.1%

The BriefThe theme today is that top-tier AI is getting cheaper and more widely available at the exact moment governments are learning how to police it. Anthropic's most powerful models (Fable 5 and Mythos 5) came back online worldwide on July 1 after an 18-day US export freeze that followed a discovered security hole, and the company launched Sonnet 5, a mid-priced model that gets close to its best one for a fraction of the cost. For anyone using these tools at work, the practical takeaway is that the affordable option keeps getting good enough for real tasks, so you rarely need to reach for the biggest, priciest model.

Level UpPick one routine task you already do this week (triaging email, turning meeting notes into action items, a first-draft script) and hand it to a cheaper model like Claude Sonnet 5 with full context: the goal, what you already tried, and the constraints. The new lower prices make it almost free to see where AI already saves you time and where it still does not. Claude Sonnet 5

Caught Up?
  • Anthropic launched Claude Sonnet 5, a cheaper mid-tier model that nearly matches its flagship, now the default for free and paid users.
  • The US lifted its 18-day export freeze on Anthropic's top models (Fable 5, Mythos 5), which return worldwide July 1.
  • Anthropic also released Claude Science, a workbench built to help researchers run reproducible, auditable experiments.
  • OpenAI's new GPT-5.6 family stayed locked to about 20 government-approved partners instead of a public launch.
  • Google shipped Nano Banana 2 Lite, a fast, cheap image generator that keeps characters consistent and renders readable text.