Meituan releases and open-sources trillion-parameter model LongCat-2.0, completing full training on 50,000 domestic chips
Original title:国产芯片上跑出的万亿参数模型,美团 LongCat-2.0 正式发布
AI summary
Meituan officially released and open-sourced the trillion-parameter large model LongCat-2.0, with 1.6T total parameters and about 48B average activation, completing the full pipeline from pre-training to inference on a 50,000-chip domestic compute cluster. The model natively supports a 1M ultra-long context, with pre-training data exceeding 30T tokens; the average daily failure rate during training dropped by more than 70%, training MFU improved 1.5x, and steady-state daily throughput exceeds 1T tokens/day.
Key Takeaways
- Meituan officially released and open-sourced the trillion-parameter large model LongCat-2.
- 0,with 1.
- 6T total parameters and about 48B average activation,completing the full pipeline from…
- The model natively supports a 1M ultra-long context,with pre-training data exceeding 30T tokens
- the average daily failure rate during training dropped by more than 70%,training MFU improved 1.
Quick Facts
- Published
- Updated
- Source
- 公众号:龙猫LongCat(美团)
- Category
- Model Releases
- AI score
- 77
官方披露了国产五万卡集群训练万亿参数模型的稳定性与效率细节,以及多项基准成绩,对国产算力路线有参考价值。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Meituan releases and open-sources trillion-parameter model LongCat-2.0, completing full training on 50,000 domestic chips. June 30, 2026. https://new.nat6.net/en/news/cmu1t9iwd034lrore39wcnw06/