As of: 2026-10-14 00:22 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/m/longcat-2-5-preview/

# LongCat-2.5-Preview

Meituan, LongCat family. Status: preview. Type: reasoning model. Released: 2026-09-25. Training cutoff: not published.

## How to call it

| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| OpenCode | `meituan/longcat-2.5-preview` |  | https://opencode.ai/data/meituan/longcat-2-5-preview.md |
| LongCat website (Meituan; API platform URL not verified) |  |  | https://longcat.chat/blog/longcat-2.0/ |

## Pricing

- Input: $0.30 per 1M tokens
- Cached input: $0.006 per 1M tokens
- Output: $1.20 per 1M tokens
- Unit as published: per 1M tokens (USD), as listed by OpenCode; Meituan's own price page not checked
- Source: https://opencode.ai/data/meituan/longcat-2-5-preview.md
- Last checked: 2026-10-13

## Specs

- Input: text, image
- Output: text
- Context window: 1,000,000 tokens
- Max output: 131,000 tokens
- Open weights: no
- Licence: proprietary

## What it has not seen

- No published training cutoff. Counted from its release (2026-09-25): at least 527 AI events since (84 major, 13 historic), as of 2026-10-14.
- Everything since August 2026: https://postcutoff.com/since/2026-08/

## What stands out

- Very large MoE aimed at long agentic coding sessions: About 1.6T parameters with about 48B active, 1M-token context, image input; reported as compatible with Claude Code, OpenCode, Kilo Code, OpenClaw and Hermes. No vendor benchmarks published. Source: https://stock.10jqka.com.cn/20260926/c680285995.shtml

## Notes

Official API base URL and model id on Meituan's platform not verified (third-party pages disagree); max output is OpenCode's figure (131K). Free on OpenCode in the first weeks. Predecessor LongCat-2.0 (June 30, 2026) was said to be trained on Chinese-made AI ASICs.


Model page: https://postcutoff.com/m/longcat-2-5-preview/