News Around the World — 2026-10-30

AI

• Co-Design Bench: Multimodal Collaborative Coding Evaluation for Agentic Models — Across 273 real-user application revisions, the strongest of ten tested model endpoints satisfies about 69.9% of acceptance criteria on average but fully completes only 26.4% of tasks.
• Your Benchmark Is an Empirical Measure Over Difficulty — Treating benchmarks as a measure over item difficulty, 10 of 25 popular benchmarks deviate from a uniform difficulty distribution and 8 have a maximum coverage gap ≥0.10; the paper’s BenchFill method cuts the top-three gap sum by 21–39% by adding just three items per benchmark.