← Back to live feed1 story
The natively multimodal LongCat-2.5-Preview AI model is now available on the OpenCode platform under a zero data retention policy. This trial provides free access for 2 weeks to a model featuring a 1 million token context window designed for long-horizon agents capable of operating across browsers, terminals, GUIs and spreadsheets.
The model utilizes a mixture-of-experts architecture with 1.6 trillion total parameters and 48 billion active parameters. Meituan omitted performance benchmarks for this launch, whereas the previous LongCat 2.0 version was trained on Ascend 910B chips and claimed state-of-the-art results. Some developers who tested LongCat 2.0 in small-scale evaluations described the predecessor as completely hopeless.
Sign in to suggest edits
Key sources
- SOURCE@maxforai“1.6T 总参数，约 48B 激活，100 万 Token 上下文，原生多模态，而且这次明显是冲着长流程 Agent 去的。”x.com
- SOURCE@opencode“LongCat-2.5-Preview is now free on OpenCode for two weeks”x.com
- SUPPORT@teortaxestex“LongCat 2.0 was the first truly large model trained on Ascends (910B to boot!), and boasted of benchmarks on par with contemporary SoTA. In the little tests I gave it, it was completely hopeless.”x.com
- SOURCEmarketbrief.now