Gemma 4, Google Search, Codex и Hermes Desktop

Свежий выпуск о Gemma 4 12B, Ideogram 4.0, AI-поиске Google, политике frontier AI, GPT-Rosalind, расходах на coding agents, Suno, Hermes Desktop и новых агентских бенчмарках.

1. Google DeepMind выпустила Gemma 4 12B — encoder-free multimodal open model runs text, image, and audio on 16GB laptops
2. Ideogram 4.0 вышла как open-weight image model — open-weight 2K image model raises the bar for text rendering and controllable layouts
3. Google дал сайтам opt-out от AI search — Search Console opt-out exposes publisher dependence on AI-shaped search traffic
4. Белый дом выпустил AI cybersecurity order — voluntary model safety testing pairs with rapid government AI cyber-defense mandates
5. Perplexity анонсировала hybrid local/cloud orchestrator — orchestrator routes tasks between local and cloud models, making privacy a scheduling problem
6. OpenAI расширила GPT-Rosalind — follow-up: life-science model adds biological reasoning, medicinal chemistry, genomics, and workflow capabilities
7. OpenAI предложила blueprint для frontier AI governance — frontier safety blueprint reframes model governance as federal resilience and national-security plumbing
8. Wasmer использовал Codex для Node.js runtime на edge — case study claims Codex accelerated a Node.js edge runtime by 10x to 20x
9. Uber ограничивает Claude Code из-за расходов — follow-up: enterprise coding-agent adoption runs into budget caps and token governance
10. Suno подняла $400M при оценке $5.4B — AI music funding doubles while copyright litigation remains unresolved
11. Nous выпустила Hermes Desktop — open-source desktop shell moves agent workflows from terminal ritual to cross-platform app
12. AutoLab проверяет long-horizon AI research — benchmark evaluates sustained iterative research and engineering rather than single-turn answers
13. StreamMA снижает latency multi-agent reasoning — streaming intermediate reasoning between agents turns pipeline depth into lower-latency cooperation
14. M3Eval тестирует память мультимодальных моделей — video benchmark asks what multimodal models retain, forget, and corrupt under interference