Little Red Book Open Dots3-note: Let's get Agent on the side and focus on the super-long mission

dots 3-note, the first open weight model of the dots 3 series, was opened at the red book dots laboratory. the total parameter of the model is 280b, with 16b activated each time, supporting 512k context, capable of processing text, pictures, video and audio, with a focus on long-range anent. models can explore rules, call tools, write codes in unfamiliar environments, and store important information in memory and adapt action to new circumstances. its training methodology, thetempo, is designed to address the problem of late feedback on long missions by splitting up a dozen hours of missions into multiple phases and allowing the model to continue its intensive learning through self-assessment. official experiments have shown that ordinary grpo tends to stagnate at a later stage, while tempo can continue to rise. the team also opened up vibesearchbench and vibelifebench, which consists of 200 missions, 22 simulation services and 288 tool interfaces, with the current average performance of the strongest model at 32.5. long-range agent still has work to do, and there is still a long way to go before it is truly a long-term mission。