Meituan releases LongCat-2.5 official version, opening online trial and API
BIBIBI
AT A GLANCE
Meituan releases LongCat-2.5 official version, saying its programming and complex Agent task capabilities have improved, and opening online trial and API。
Article
Meituan LongCat-2.5 officially releases: Coding capabilities significantly improved, with benchmark scores approaching Opus 5 Dongcha Beating AI Flash News: Meituan releases LongCat-2.5 the official version, more than a month after this year's September preview version went live.
The model retains the previous generation's 1600000000000 total parameters, approximately 48000000000 active parameters, and 1000000 Token context window, with a focus on improving visual programming and complex Agent task capabilities.
In official testing, the Terminal Bench v2.1 score rose from the previous generation's 70.8%to 88.4%, approaching Claude Opus 5's 89.1%. On the more difficult v4.0, it also increased from 3.0%to 38.5%, but still trails Opus 5's 53.9%.
The official version is now available for online trial and API, while the model weights have not yet been open-sourced. The limited-time price is Token per million uncached input 2 yuan and output 8 yuan.
Original link https://m.theblockbeats.info/flash/375021
Key points
01
Dongcha Beating AI says that Meituan has released the LongCat-2.5 official version.
02
The report states that the official version came more than a month after this year's September preview version went live.
03
The model retains 1600000000000 total parameters, approximately 48000000000 active parameters, and 1000000 Token context window.
04
The report says that the model focuses on improving visual programming and complex Agent task capabilities.
05
In official testing, the Terminal Bench v2.1 score rose from 70.8%to 88.4%，Claude Opus 5 was 89.1%。
06
On Terminal Bench v4.0, the score increased from 3.0%to 38.5%，Claude Opus 5 was 53.9%。
07
The official version is now available for online trial and API, while the model weights have not yet been open-sourced.
08
The limited-time price is Token per million uncached input 2 yuan and output 8 yuan.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
LongCat-2.5 is the official version of a model released by Meituan. The report emphasizes its programming, visual programming, and complex Agent task performance, and compares Terminal Bench scores with the previous generation and Claude Opus 5. It is currently available for online trial or through API use, but the model weights are not available.
Why it matters to readers
If the capabilities, scope of availability, and pricing in the report are further confirmed, developers will have another model option available for online trial and through API calls; however, the current performance figures come from official testing.
Beginners can follow the online trial and API, but benchmark scores do not equate to performance in all real-world use cases, and limited-time pricing does not represent long-term pricing.
Risks and unknowns
The provided material does not include Meituan's original announcement or technical report; the model specifications, scope of availability, and pricing still need to be verified.
The performance scores are described as official testing, with no independent evaluation results provided.
It does not specify the regions of availability, actual stability, or usage restrictions for the online trial and API.
It does not disclose when the limited-time pricing ends or the subsequent pricing.
The headline says benchmark scores approach Opus 5, but the v4.0 score still trails Claude Opus 5。
Related Developments
Loading event timeline…
Related concepts
Token
This term is not in the glossary yet. Browse related concepts in the glossary.
