Cursor released results from an upgraded agent swarm architecture that separates planner models from worker models, asking the system to rebuild SQLite in Rust using only documentation with no source code or internet access. Every configuration of the new system achieved 100 percent on the test suite, suggesting that cheaper inference models can execute coding work effectively when frontier models perform planning (The Decoder).
Black Forest Labs released FLUX 3, a multimodal flow model that learns from images, videos, and audio within a single architecture and ships video, audio, and robot action prediction from one set of weights (MarkTechPost). In parallel, the KwaiKAT team at Kuaishou published KAT-Coder-V2.5, an agentic coding model trained on over 100,000 verifiable repository environments, arguing that agentic capability is bottlenecked by training infrastructure rather than model scale; their AutoBuilder system raised environment construction success from 16.5 percent to 57.2 percent (MarkTechPost).