Skip to content
NewA calmer, faster ZeroHour

Meituan releases a 1.6-trillion-parameter agent model trained without Nvidia GPUs

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

The Decoderby Jonathan Kemper ()3 min read1
AI summary · x-ai/grok-4.7Summaries by Vani AI

MIT-licensed LongCat-2.5 activates about 48 billion parameters per token; Meituan scores it 88.4 on Terminal Bench v2.1 versus Opus 5 at 89.1.

Chinese food-delivery company Meituan published weights for LongCat-2.5 one week after unveiling it. The mixture-of-experts model targets long autonomous agent work, especially coding, and was trained on Chinese chips rather than Nvidia GPUs. Its API accepts up to one million tokens and can emit 128,000, with text, image, and video input. A sample self-hosting guide uses 194 Huawei Ascend Atlas A2 accelerators and gives no Nvidia setup.

Unconfirmed: Benchmark standings and the claim of parity with the uncompressed model are Meituan's results; BrowseComp was run in Meituan's environment.

Key facts
  • A 4-bit build raised prompt throughput 43.3 percent versus 8-bit in Meituan's tests.
  • Terminal Bench v4.0 score is 38.5, below Claude Opus 5's 53.9.
  • Promotional pricing is about €0.27 per million input tokens and €1.07 per million output tokens.
  • Meituan says predecessor LongCat-2.0 was trained in late June on more than 50,000 Chinese chips.
  • The API supports OpenAI and Anthropic formats and plugs into Claude Code, Codex, and Hermes.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

From the original article

3 min read at the-decoder.com

Meituan has released the weights for LongCat-2. 5, a 1. 6-trillion-parameter model built for AI agents and trained entirely on Chinese hardware.

The Decoder

By Jonathan Kemper · the-decoder.com

Read the full article at The Decoder

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.