by Meituan
Meituan's open-weights 1.6T-parameter multimodal MoE model (48B active) for agentic coding and visual tasks.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
LongCat-2.5 is a native multimodal mixture-of-experts model from Meituan's LongCat team, with 1.6 trillion total parameters and roughly 48 billion activated per token. It builds on LongCat-2.0 and accepts image and text input, producing text output. It is aimed at long-horizon agentic tasks such as complex codebase development, full-project builds, creative design and 3D-related tasks.
According to the model card, visual features at native resolution are converted into discrete tokens and aligned with text in a shared space (building on LongCat-Next). Training used large-scale reinforcement learning on real-world engineering tasks, multi-teacher distillation, and long-sequence training optimizations run on domestic AI accelerators. Benchmark results in the card are partly cited from public sources and partly measured in-house, so they are indicative only.
Weights are published on Hugging Face under the MIT license. Creators noted the model is also reachable via LongCat Chat and an API. At 1.6T parameters it is too large for most people to run locally.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open 1.6T parameter multimodal model with 48B active parameters, with stronger visual understanding and strength in creative design and 3D tasks. Too large for most people to run locally.
Multimodal 1.6T MoE (48B active) strong at coding, visual understanding and autonomous tasks. Available via LongCat Chat and API; open-source release marked coming soon.
Closest neighbours by meaning, not just by category.
Model · Meituan
Meituan's open-weights 1.6T-parameter mixture-of-experts LLM (48B active), aimed at agentic coding and reasoning tasks.
Model · MiniMax
Open-weights native multimodal MoE model from MiniMax, ~428B parameters (~23B active) with 1M-token context.
Model · Thinking Machines
Open-weights multimodal MoE model (276B total, 12B active) from Thinking Machines; text, image and audio in, text out.
Model · IQuest Lab
320B-parameter MoE model (15B active) from IQuest for agentic coding and tool use, with 512K context.
Model · Z.ai
Z.ai's open-weights 753B MoE language model, a post-trained GLM-5.2 aimed at coding, agentic tasks and cybersecurity.
Model
Open model family (9B dense, 35B MoE, 397B MoE) trained with a self-improvement loop for reasoning, coding and agentic tasks.