ai/intel

Read

  • Today
  • Archive
  • Briefs
  • Labs

You

  • Settings

Live · updated 6:27 AM

ai/intel
Today
Today / Chips / Story

Meituan releases a 1.6-trillion-parameter agent model that never touched an Nvidia GPU

Sun 15 Nov · 10:06 AM, updated 10:06 AMFirst seen Sun 15 Nov · 10:06 AMUpdated 10:06 AMLimited coverage, tapering
Nvidia · ChipsNVDA$125.53▼10.9%ANT$88.91▼9.7%

Reported by The Decoder1 source ↓

Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter agent model trained entirely on Chinese hardware. It scored 88.4 on Terminal Bench v2.1, just below Claude Opus 5 in Meituan’s comparison.

Key points

  • The model has 48 billion active parameters per token
  • Meituan’s self-hosting guide specifies a cluster of 194 Huawei Ascend Atlas A2 accelerators
  • The weights are available under the MIT license for commercial use, modification and redistribution
  • The API supports up to 1 million input tokens and responses up to 128,000 tokens
  • Promotional API pricing is about €0.27 per million input tokens and €1.07 per million output tokens

Sources

  1. The Decoder10:06 AMRead ↗Meituan releases a 1.6-trillion-parameter agent model that never touched an Nvidia GPU

Details

The Decoder reports that Meituan released the weights one week after LongCat-2.5’s official unveiling. The model is designed for autonomous, multi-step agent tasks, particularly coding, and can process text, images and video while returning text.

LongCat-2.5 is a mixture-of-experts model with 1.6 trillion parameters in total and about 48 billion active for each token. Its API supports context windows of up to 1 million tokens and outputs of up to 128,000 tokens. Meituan’s self-hosting guide describes a 194-accelerator Huawei Ascend Atlas A2 cluster; the article reports no Nvidia GPU guidance.

Meituan reports a score of 88.4 on Terminal Bench v2.1, compared with 89.1 for Claude Opus 5 in its table. The weights are released under the MIT license, and the API supports OpenAI and Anthropic formats. The Decoder reports promotional prices of roughly €0.27 per million input tokens and €1.07 per million output tokens.

Written from coverage as of 10:17 AM

Images

  • The Decoder
  • The Decoder
  • The Decoder

Image via The Decoder(opens the article in a new tab)

From The Decoder
1 of 3

Stocks in this story

Change since open: 20 Nov to 21 Nov 2026

NVDANvidia, stories

$125.53

Down14.0%

After hours$130.01Up3.6%

ANTAnthropic, stories

$88.91

Down8.5%

After hours$87.79Down1.3%

Last refresh 00:00

More on Nvidia

  • Jensen Huang Resigns as NVIDIA CEO; CFO Colette Kress Named Interim SuccessorWed 11 Nov · 68 outlets
  • Nvidia board reportedly prepares to oust Jensen Huang after second affair reportThu 5 Nov · 51 outlets
  • Nvidia revenue doubles to $115.8 billion in first results after Huang exitWed 11 Nov · 33 outlets
  • Nvidia board meets to discuss Jensen Huang’s future as CEOWed 11 Nov · 27 outlets