Hey hustlers,
The government banned one of the most powerful AI models on the planet. Then un-banned it. In 18 days. With no real playbook to follow.
Meanwhile, new models dropped, old ones got cheaper, and Google quietly released two image generators while nobody was watching.
A lot happened this week.
Let's take a look.
👾 WHAT'S NEW IN AI
1. Fable 5 is back. The 18-day government ban is over.
On June 12, the US Commerce Department applied emergency export controls to Claude Fable 5 and Mythos 5, citing cybersecurity risks after a jailbreak was discovered. Anthropic had no way to verify user nationality in real time, so they shut both models down for everyone globally. On June 30, Commerce Secretary Howard Lutnick withdrew the order. Fable 5 came back July 1 across Claude.ai, Claude Code, and Claude Cowork.
Why you should care: This is the first time a frontier AI model pulled by government order has been walked back online. No formal process existed for any of this. Washington improvised. The bigger story is what it signals going forward: frontier model launches are starting to look less like product releases and more like negotiated deployments. That changes things for every major AI lab.
2. Claude Sonnet 5 is now the default for everyone
Anthropic launched Claude Sonnet 5 on June 30 and made it the default model for every Free and Pro user starting July 1. It performs close to flagship Opus levels on most tasks and comes in at introductory API pricing of $2 input / $10 output per million tokens until August 31.
Why you should care: Enterprises had been burning through AI budgets fast in Q2 2026, as agentic workflows ate through tokens at an alarming rate. Sonnet 5 is Anthropic's direct answer to that. Frontier-adjacent agents, for a fraction of the cost. This is the model that will define how most people use Claude for the next few months.
3. Google dropped two new image generation models
On June 30, Google released Gemini 3.1 Flash Image at $0.50 input per million tokens and Gemini 3 Pro Image at $2.00 input, both immediately available in Google AI Studio and the Gemini API.
Why you should care: These launches exist because Gemini 3.5 Pro keeps missing its release window. Google needed to fill the lineup and stay relevant heading into Q3. For you, this means solid image generation at genuinely low cost right now, with both models available to test today. Flash for volume. Pro for quality.
👾 THE GOOD STUFF
🔧 AI Tool: Itential FlowAI
An agentic operations platform built specifically for infrastructure teams.
You describe what you need, it generates and deploys governed
workflows through your existing APIs.
Lumen cut enterprise activation from 45 days to under one using it.
🐙 GitHub: mem0ai/mem0 (59k+ stars)
The universal memory layer for AI agents.
Gives your agents the ability to remember users, adapt over time, and
build actual context across conversations. 58k stars and growing fast.
🎬 YouTube: Claude Science FIRST LOOK
A hands-on look at Anthropic's new Claude Science app, showing exactly
how it handles protein visualization in real time.
Worth watching if you're curious about where AI meets research.
👾 TO READ
Orca: The World is in Your Mind
📎 https://arxiv.org/abs/2604.24618
Researchers at the Beijing Academy of Artificial Intelligence asked a question that sounds simple but isn't: what if one AI model could understand the world the way humans do, across video, language, and physical action, all at once? Instead of training separate models to predict the next word, the next frame, or the next move, they built Orca around a single idea called Next-State-Prediction. One model, one way of understanding how things change.
What they found: Pre-trained on 125,000 hours of video and 160 million event annotations, a single frozen Orca model outperformed specialized baselines on text generation, image prediction, and physical action tasks without any task-specific fine-tuning. The same underlying model. Three completely different outputs.
Why it's interesting: This week Anthropic brought back Fable 5 and launched Sonnet 5, both heavily agentic models designed to take real actions in the real world. For agents to act well, they need to understand cause and effect, not just predict the next token. Orca is an early blueprint for how that might actually work. The gap between AI that talks and AI that understands is exactly what this paper is trying to close.
🧵 Thread Drop
This week felt like a turning point. A government that usually moves slowly made a snap decision on AI, reversed it 18 days later, and admitted there was no real process in place. That's not a small thing.
The models are getting cheaper, faster, and more capable. The rules around them are still being written in real time.
👾 See you soon 👾
Want to work together?
If you're building something in AI and
think our audiences would vibe with it, let's talk.
