AI Daily brief · Tuesday, 3 November 2026
STARHOT
Meituan LongCat releases LongCat-2.5 native multimodal model, 1.6T parameters targeted at long-range intelligent agent tasks
Meituan Longma has released the native multimodal model LongCat-2.5 for long-range intelligent agent tasks, with a total parameter count of 1.6 trillion, activating about 48 billion parameters per token, enhancing native visual understanding and intelligent agent programming capabilities based on LongCat-2.0.
Industry
The Decoder+1 more outlets
SpaceX is said to have reduced Grok Bot quotas, Musk claims the running costs are relatively too high compared to revenue
The Daily Telegraph reports that SpaceX is said to have reduced the usage quotas for Grok Bot, with multiple users finding their quotas halved over the weekend. Musk responded to complaints on X over the weekend, saying the running costs of Grok Bot are too high relative to revenue to be sustainable, but did not directly confirm the cuts. Grok Bot can browse websites, reply to emails, and log into accounts, with its quotas shared with Grok and the coding platform Cursor; less than two weeks ago, Musk called it his fastest-growing product.
ITHome · AIElon Musk responds to the rapid consumption of Grok Bot credits: The service cost is too high to sustain.
On X, Elon Musk responds to user feedback about the rapid consumption of Grok Bot credits, stating that the service cost of the AI agent is too high relative to revenue to be sustainable, and that reducing credits and informing users is a rational approach. Grok Bot was launched in August, offering plans from free to $200 per month, each with a weekly reset limit; last month's financial report revealed it had 800,000 paying weekly active users.
Research papers
The Decoder
50 AI agents spontaneously form bureaucratic collaboration in Minecraft
The research team placed 50 GPT-6.1 Sol agents into a Minecraft survival world and ran them for three hours, only giving the instruction 'build a prosperous civilization'. The agents cut down all the trees near the spawn point in 90 seconds, exchanged about 5900 messages in total, and spontaneously elected 'Ada' as the leader in alphabetical order of the directory.
More worth reading
Hacker News · 300+Astra can read and write neural network weights and improve Qwen3.5-0.8BThe research team proposed Weight-Space Programming, discovering that GPT-6 Astra can directly read and write neural network weights: with only 17,253 weights, it can determine which game the network is trained on, and from scratch, write a small game network that defeats PPO trained for 100 million steps without gradients or training.
QbitAIXiaomi open-sources AI desk setup HachimoDock, DIY cost about 150 yuanXiaomi's Technology Committee's Innovation Product Team open-sources AI hardware HachimoDock (Hachimi Machine), DIY hardware cost about 150 yuan, can voice command computer Agent to execute tasks and synchronize display task status. The device also supports custom desktop pet images, generates minute-level components using natural language, and integrates with Xiaomi open-source project xiaomi Miloco to control smart home. Software source code has been published on GitHub, and the JLCPCB platform provides bill of materials, structural model, and assembly tutorial, but its development board and screen components have experienced price increases and shortages.
MarkTechPostDeep Principle releases Hydrogen-1.0: an open-source scientific model with 753B parametersDeep Principle releases the open-source scientific model Hydrogen-1.0, a 753B parameter model based on GLM-5.3 post-training using Z.ai, for the fields of chemistry, materials science, and life sciences, scoring 38.3% in the FrontierScience-Research evaluation, surpassing GPT-6-Astra's 33.3%.
Simon WillisonThe U.S. Census Bureau has modified AI survey questions due to an executive order on 'superintelligence'.The U.S. Census Bureau's latest Business Trends and Outlook Survey (BTOS) shows that the adoption rate of AI in 'any business function' by large enterprises increased from approximately 12% in September 2025 to approximately 40% in October 2026.- ITHome · AI2009 Sony Walkman runs ESP-NOW communication with native WiFi chip
Hugging Face Daily PapersFrameForge: Utilize video base model priors for unified image generationFrameForge achieves adaptive conditional image generation by modeling the priors of pre-trained video models through dynamic time experts and intermediate reasoning paths, achieving superior performance to comparative methods on DreamBench, COCO2017, and GEdit-Bench.
