AI Brief

Zyphra released the ZAYA1-8B 8B MoE model with 760M active parameters, matching top models on math benchmarks.

Zyphra · May 7, 2026foundation model12 signals

Google launched Google Cloud Fraud Defense, a platform that verifies bots, humans, and AI agents as the next evolution of reCAPTCHA.

Google · May 6, 2026safety329 signals

Google introduced Multi-Token Prediction drafters for Gemma 4 models to speed inference via speculative decoding.

Google · May 6, 2026foundation modelT1669 signals

Anthropic introduced 'Dreaming', a feature that analyzes AI agent activity logs to extract insights and improve agent performance.

Anthropic · May 6, 2026agent frameworkT1115 signals

Anthropic raises usage limits for Claude Code and Claude API, doubling rate limits and removing peak hour caps.

Anthropic · May 6, 2026foundation modelT11,078 signals