Anthropic launches "Dreaming" feature for Claude Managed Agents—agents automatically organize memories and identify patterns during idle time, boosting efficiency up to 6x. This is the industry's first systematic agent memory management solution, alongside multi-agent collaboration and outcome assessment capabilities.

Musk announces xAI dissolution into SpaceX, with the world's largest supercomputer Colossus 1 (220K GPUs) fully rented to Anthropic at $5B/year. xAI projected annual loss of $13B with Grok utilization at only 11%. Transition from model competitor to compute supplier.

Moonshot AI closes ~$2B Series D, post-money valuation exceeds $20B, setting a record for Chinese LLM funding. April launch of open-source flagship Kimi K2.6 achieves 1.85T weekly tokens on OpenRouter, ranking second globally. Focused on Agent赛道, next-gen K3 aims to match frontier performance.

2026 Agentic AI Readiness Index reveals: nearly 60% of enterprises have invested millions, but only 15% have data infrastructure ready for safe agent scaling. Data quality & lineage (42%), compliance (39%), security & privacy (39%) are top bottlenecks. Infrastructure becomes the ROI limiter.

ByteDance's Doubao (345M MAU) announces paid tiers: ¥68 Standard, ¥200 Enhanced, ¥500 Professional. Free service retained, paid tiers focus on complex productivity scenarios. Chinese LLM industry shifts from burning cash for traffic to value monetization.

Meta, Stanford, Harvard jointly release new benchmark: 200 real software projects requiring full rewrite from scratch. All 9 top models achieve 0% complete success rate, best performer Claude Opus 4.7 averages only 51.2%. A brutal stress test for AI coding capabilities.