Mid-2026 field brief of popular frontier, open-weight, and IDE-native models — Claude, GPT-5.6, Gemini, Grok, Composer, DeepSeek/GLM/Qwen — and how to pick by task, cost, and control.