OpenCode Go
LongCat 2.5 is Meituan's newest model from its LongCat team, published with a tech blog on November 1, 2026 and generally available on the LongCat API platform on November 2 after a September 25 preview. It is a natively multimodal mixture-of-experts model with 1.6 trillion total parameters and roughly 48 billion active per token, taking text, images and video as input and producing text. The model keeps LongCat 2.0's text-only 1.6T MoE backbone but adds image input by turning pictures into multi-level discrete tokens, so one stream handles both text and pictures. It reports a 1,048,576-token context window with up to 262,144 output tokens, runs on Chinese domestic accelerators with W4A4 inference, and ships Anthropic-format and OpenAI-format endpoints with guides for Claude Code, Codex, OpenCode and Cline.
