Google released Gemma 4 on April 2, 2026 under Apache 2.0 — their first fully permissive open model. The 31B dense model scores 89.2% on AIME and 80.0% on LiveCodeBench. The 26B MoE variant delivers 97% of that performance with only ~4B active parameters. Here's the full breakdown.
Z.ai (formerly Zhipu AI) releases GLM-5.1, scoring 45.3 on coding benchmarks — just 2.6 points behind Claude Opus 4.6. Trained entirely on Huawei chips, open-source under MIT license, and starting at $3/month. Here's what the benchmarks actually show and what's still unverified.
A landmark study of 9,649 experiments across 11 models reveals why context engineering — not prompt engineering — is the critical AI skill in 2026. Frontier models beat open-source by 21 points, and how you structure context matters more than how you write prompts.
