Meituan releases LongCat-2.5-Preview with image understanding

Summary

Meituan opened access to the LongCat-2.5-Preview model via API on September 25, 2026. The model offers visual understanding capabilities (visual question-answering, summarization, visual reasoning) while ensuring compatibility with tools like Claude Code and OpenCode. The company also claims improved coding performance.

Why it matters

Meituan, primarily known as a food delivery and local services giant, is now competing in the foundation model space. The addition of multimodal (visual) understanding capabilities signals the company's effort to move beyond text-based large language models (LLMs). Explicit compatibility with Western developer toolchains like Claude Code and OpenCode indicates that Meituan is targeting not only domestic but also international developer audiences. This further expands an already crowded field of Chinese-origin open-weight and API-based models that developers can integrate into their existing workflows.

Highlights

• Preview-stage release: The article does not specify benchmark results, parameter count, or context window specifications.
• Visual understanding covers question-answering, summarization, and 'complex visual reasoning.' However, no concrete examples or evaluation scores are provided.
• Compatibility with Claude Code, OpenCode, and OpenClaw is highlighted as a differentiator for developer adoption.

Critical take

The article is a weak repetition of Meituan's own documentation, lacking independent verification. It omits benchmark results, sample outputs, limitations, or pricing information. The claim of 'improved coding performance' is entirely unquantified. While the headline presents it as a full product launch, the maturity of what the company itself describes as a 'preview' version under development is overstated.