Morning brief

Wednesday, July 1

1 people · 0 companies · 5 topics

TL;DR Fable 5 is back globally after 2.5 weeks offline, jailbreak patched and US export controls lifted June 30. Anthropic also shipped Claude Sonnet 5 and Claude Science on June 30, making it the biggest Anthropic drop in months.

Models & research

  • Claude Fable 5 and Mythos 5 restored globally as of July 1 across Claude.ai, Claude Code, API, and Cowork. US Commerce lifted export controls June 30. Anthropic’s new classifier blocks the Amazon-found jailbreak in 99%+ of cases; trade-off is a higher false-positive rate on coding requests that now fall back to Opus 4.8 with user notification. Usage capped at 50% of weekly limit through July 7 while Anthropic ramps. AWS/GCP/Foundry reinstatement pending. The Hacker News · The Decoder

  • Claude Sonnet 5 launched June 30. Near-Opus-4.8 quality; now default for Free, Pro, and claude-code. Pricing: $2/$10 per Mtok intro through Aug 31, then $3/$15. 1M token context. One catch: the new tokenizer generates ~30% more tokens on the same input versus Sonnet 4.6, so effective costs may run higher than sticker. Anthropic · TechCrunch

  • OpenAI released GeneBench-Pro (June 30): 129 synthetic biology/genomics problems judging AI agent performance in realistic research workflows. GPT-5.6 Sol Pro tops at 31.5% pass rate; Claude Opus 4.8 at 16%. Top models fail 70% of realistic biology tasks. 10 problems open-sourced on Hugging Face. OpenAI · AlphaSignal

People

  • emollick published “The Twilight of the Chatbots” (June 30): argues the chatbot era is over and agentic systems are now coming to work beyond just engineering. Cites METR and UK government estimates of programmer-hours-per-prompt as the key capability metric to watch. oneusefulthing.org

Tools & dev

  • Claude Science launched in beta June 30: a dedicated research workbench for scientists on Pro/Max/Team/Enterprise (macOS and Linux). Pre-configured with 60+ scientific databases, renders 3D protein structures and genome browser tracks, and runs a reviewer agent that checks citations and math end to end. Early users include Novo Nordisk and Allen Institute. Anthropic funding up to 50 projects at $30k in credits each; applications open through July 15. CNBC · TechCrunch

OSS

  • msitarzewski/agency-agents is today’s trending standout (2,097 stars): 112 specialized AI agent personas installable into Claude Code, Cursor, Codex, and Gemini via a native macOS/Linux/Windows app. MIT licensed, local-first, no telemetry. GitHub

  • virgiliojr94/book-to-skill trending today (205 Python stars): turns technical PDFs into Claude Code skills automatically. GitHub

  • langchain-ai/deepagents trending today (103 Python stars): LangChain’s batteries-included agent harness. GitHub