Claude includes Fable 5 in Max & Team Premium

#1 𝕏

Claude will include Fable 5 in all Max and Team Premium plans at 50% of normal limits starting July 20. Pro and Team Standard users keep access via usage credits (with a one-time $100 credit), with the staged rollout helping manage unpredictable demand.

Also covered by: @Simon Willison

#2 📝 Anthropic Engineering

How we contain Claude across products - Anthropic deploys three agentic products—claude.ai, Claude Code, and Claude Cowork—using two containment strategies: human-in-the-loop approvals (users approved roughly 93% of permission prompts) and environmental containment (sandboxes, VMs, filesystem and egress controls); Claude Code’s auto mode automates safer approvals and catches roughly 83% of overeager behaviors before they execute. Anthropic withheld Claude Mythos Preview in April 2026 for having too high a blast radius and reports that model-layer defenses are strong but imperfect—Claude Opus 4.7 holds prompt-injection attack success to about 0.1% on single attempts and roughly 5–6% after 100 adaptive attempts—so overlapping defenses across environment, model, and external content are required.

#3 in

Peter Yang built a ChatGPT-powered site with his 8-year-old to teach multiplication tables, using ChatGPT Images for animal-themed UI and characters, plus added music and a timed boss level.

#4 𝕏

Thariq has put in extensive work to ensure Fable remains continuously available on MAX subscription plans, driven by overwhelming user demand.

#5 📝 Ampcode Chronicle

Subscriptions, At Last - Amp Subscriptions launches in Beta with two monthly plans: Megawatt at $20/month (all product features, 750 hours of small orbs, $20 included agent usage, low & medium modes only; use your ChatGPT sub for GPT-5.6 and 𝕏 Premium+/SuperGrok) and Gigawatt at $200/month (everything in Megawatt plus 1,000 hours of large orbs, $200 included agent usage, and all modes including high & ultra). Subscribers who sign up on or before Sunday, July 191 get bonus 2x usage for orbs and agents in the first month; Amp still supports pay-as-you-go, but to exceed subscription limits you must link your ChatGPT subscription or add paid credits, and the company emphasizes agent work running in remote "orbs" rather than local dev environments.

#6 𝕏

Guillermo Rauch ran Kimi K3 on Vercel’s private DeepSec cybersecurity benchmark and found it outperformed every GPT-based competitor except Sol—which delivered similar results at 7× the cost.

#7 𝕏

Jason Zhou is rolling out an internal “graph of loops” framework that shares state and artifacts across iterative processes so loops compound rather than conflict. He’ll post a detailed breakdown of learnings in the next few days.

#8 𝕏

Yann LeCun argues that the openness and “ungovernability” that drove Linux and the Internet to success will likewise be the crucial factor behind open-weight AI foundation models.

#9 𝕏

Jeff Dean built tightly connected clusters of TPUs he calls “pods,” linked by a custom high-speed network. The very first TPU was codenamed Jellyfish, but since multiple jellyfish form a “smack,” he opted for the more fitting name “pod.”

#10 in

Udi Menkes cautions that easy AI prototyping has turned “you can build it in two days” into a dangerous mantra that fuels overbuilding. The true PM signal now lies in the trade-offs, choices, and disciplined experimentation behind each feature—not just shipping code.

#11 𝕏

Yann LeCun clarifies that governments worldwide require AI sovereignty rather than building their own digital public infrastructure. He predicts that open-weight foundation models will inevitably become the dominant form of AI.

#12 𝕏

Claude guaranteed Fable access at 50% usage for their most intensive plans to give customers more certainty, and they’re investing in new capacity with regular updates.

Also covered by: @Simon Willison

#13 𝕏

Thariq confirms they’ve always aimed to include it as a regular, capacity-based subscription feature.