Meituan LongCat releases LongCat-2.5 multimodal model
11/02/2026 — 11/02, 12:19·2 sources (all)·2 reports (all)·In the 2026-11-02 briefing
Key takeaways
- Meituan's LongCat team released LongCat-2.5, a 1.6T-parameter model with about 48B active parameters per token. Building on LongCat-2.0, it…
Background & analysis
On November 2, 2026, Meituan's LongMao team released LongCat-2.5, a new large model with 1.6 trillion total parameters and roughly 48 billion parameters activated per token. Building on LongCat-2.0, the model strengthens native visual understanding and agentic coding capabilities, while supporting long-context sequences and long-horizon reinforcement learning training.
Later the same day, 36Kr reported additional details: LongCat-2.5 is an原生 multimodal model aimed at long-horizon agentic tasks. It continues the research direction of the earlier LongCat series and was trained entirely on domestic Chinese computing infrastructure, which the company describes as the largest training job completed on domestic compute to date. According to the evaluation results cited, LongCat-2.5 scores between Claude-Opus-4.8 and Claude-Opus-5 on a range of mainstream benchmarks, and surpasses Claude-Opus-5 on some of them. So far, public information covers only the model's architecture and training background; no full technical report or deployment details have been released.
AI-generated from 2 reports · updated 1 hour ago
Latest turnMeituan released LongCat-2.5 on November 2, a natively multimodal model with 1.6T total parameters aimed at long-horizon agent tasks, according to 36Kr. The company says it was trained entirely on domestic Chinese compute, the largest such training run to date, and reports benchmark scores landing between Claude-Opus-4.8 and Claude-Opus-5, beating the latter on some evaluations.
- Heat index
- 187
- All sources
- 2
- All reports
- 2
- First seen
- 3 hours ago
Reports on this story headlines open the original
Meituan released LongCat-2.5 on November 2, a natively multimodal model with 1.6T total parameters aimed at long-horizon agent tasks, according to 36Kr. The company says it was trained entirely on domestic Chinese compute, the largest such training run to date, and reports benchmark scores landing between Claude-Opus-4.8 and Claude-Opus-5, beating the latter on some evaluations.
36氪AI score 67Meituan's LongCat team has released LongCat-2.5, a 1.6T-parameter model that activates about 48B parameters per token. Building on LongCat-2.0, it strengthens native visual understanding and agentic coding, with support for long sequences and long-horizon reinforcement learning training.
掘金AI score 64
Other stories people are talking about
- Heat index 705RisingDeepSeek open-sources dsh-superagent coding agent prototype12 sources
- Heat index 389OpenAI trims funding target; Anthropic may get BB rating at IPO11 sources
- Heat index 283SurgeAnthropic renames Claude Max 20x to Ultra and drops the 5-hour session cap4 sources
- Heat index 266SurgeOpenAI doubles ChatGPT Pro tier allowances, reverses Pro 200 cut5 sources
- Heat index 191NewOpenAI to merge ChatGPT and Codex usage into one shared quota from December2 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion