Anthropic released Claude Fable 5 on June 9 — its first publicly available Mythos-class model, scoring 10%+ over Opus 4.8 on some benchmarks and already selectable in Claude Code and GitHub Copilot. It is free through June 22, then roughly $10/$50 per million tokens, with cyber and biology prompts handed back to Opus 4.8.
Z.ai launched GLM 5.2 on June 13 across its Coding Plan tiers with a 1M-token context aimed at repo-scale agentic work, with MIT open weights and an API due the following week. No benchmarks at launch.
CNBC reports Microsoft and Google are pushing their own coding models into the agentic-development market, turning the coding-agent fight into a four-way race at the model layer over who powers developers' tools.
Avataar.ai's new IndiaAI-backed model Varya uses a distilled architecture that cuts video generation from 50 steps to 4 — betting that lower compute per clip, not higher fidelity, is what widens access to AI video.
Blackmagic released Resolve 21.0 on June 3, adding a Lightroom-style Photo page plus eight AI tools — de-aging, motion deblur, UltraSharpen upscaling and IntelliSearch — most free, with Magic Mask reserved for Studio.
Meta previewed a desktop Edits app and an AI assistant that reads a creator's Instagram views and retention to suggest ideas and trending audio — a direct move on CapCut. The assistant is in testing; desktop is "coming soon."