Weekly AI news roundup recorded live from the AI Engineer conference in London. Key stories include Anthropic's ARR jumping from $19B to $30B and the announcement of Claude Mythos Preview — a frontier cybersecurity model deemed too dangerous (or too compute-intensive) to release publicly, capable of finding zero-days and chaining exploits. Meta Superintelligence Labs debuted Muse Spark, a latency-optimized model built on a new stack in 9 months, with strong benchmarks and 16 hidden agentic tools discovered in the meta.ai UI. OpenAI's Codex hit 3M weekly active users (up from 2M last month), with Guardian Approvals and plugin workflows highlighted. OpenClaw 2026.4.5 shipped with a /dreaming GA feature for agent memory consolidation. Gemma 4 crossed 10M downloads. Seedance 2.0 launched in the US, jumping ~80 ELO points above competing video models on Arena.

18m read timeFrom sub.thursdai.news
Post cover image
Table of contents
Anthropic announces $30B ARR and Mythos, their next model, will not be released to the public.Meta Superintelligence Labs ships Muse Spark — Llama is dead, long live MuseGuest highlightsThis Week’s Buzz
4 Impressions