As an agent dev, which open-weight LLM is currently your default brain in 2026? Gemma 4 / GPT-OSS / Kimi K2.6 / DeepSeek V4 — or something else

seen this X poll going around about favorite open-weight AI models this year (Gemma 4 / GPT OSS / Kimi K2.6 / DeepSeek). curious what the breakdown looks like specifically for people building agents in 2026, not just general LLM use. my current stack: - Kimi K2.6 / K2.7 Code for long-horizon coding agents (the 262K context holds multi-file refactor sessions clean) - DeepSeek V4 Pro as a general fallback / cheaper tool-calling tier - GPT-OSS for stuff that genuinely needs the OpenAI style of output - haven't