Flash

The news, the context, and what changes.

RSS
Flash

Xiaomi Announces Permanent Price Cut for MiMo-V2.5 Series API

Xiaomi's MiMo-V2.5 series API announces a permanent price cut, with reductions up to 99% and Token Plan usage increased by 5 to 8 times. The trillion-token incentive plan ended early, distributing 100 trillion tokens to over 548,000 applicants. Existing user quotas are reset, and the V2.5-TTS series remains free for a limited time.

88Z AI editorial1 min read
Flash

Kuaishou Open-Sources Self-Developed Multimodal Large Model Keye-VL-2.0-30B-A3B

Kuaishou has officially open-sourced its self-developed multimodal large model, Keye-VL-2.0-30B-A3B. It introduces the DSA sparse attention mechanism to multimodal understanding for the first time, supporting a 256K ultra-long context window and enabling temporal causal reasoning for long videos. The model surpasses Gemini 2.5 Pro and Gemini 3 Flash on certain metrics in the TimeLens benchmark and unlocks Agent collaboration capabilities for complex tasks like coding, tool use, and search.

88Z AI editorial1 min read
Flash

Xiaomi Launches New World Model Xiaomi Auto World Model

Xiaomi EV has released a new framework that deeply couples 3D reconstruction with video generation. The reconstruction module uses sparse 3D anchors, processing 10-second videos in just 10 seconds. The generation module achieves efficient output of 0.19 seconds per frame with 4-step denoising via two-stage training and ODE distillation, supporting videos up to one minute long.

88Z AI editorial1 min read
Flash

Alibaba Tongyi Open-Sources Multi-Agent Development Framework AgentScope 2.0

Alibaba Tongyi Lab releases AgentScope 2.0, featuring systematic upgrades for production stability. The new version introduces model fault tolerance, real-time message visibility, permission-based interception of high-risk operations, and context optimization to enhance multi-agent reliability in real-world scenarios.

88Z AI editorial1 min read
Flash

ModelBest and Tsinghua University Open-Source Edge Text Foundation Model MiniCPM5-1B

ModelBest, together with Tsinghua University and the OpenBMB open-source community, has released MiniCPM5-1B. The model scored 17.9 on the AA-Index leaderboard, surpassing all models under 2B parameters and some larger ones, making it the strongest open-source foundation model within the 2B parameter range globally. After INT4 quantization, its size is only 0.5GB, enabling smooth operation on terminals such as mobile phones and browsers.

88Z AI editorial1 min read
Flash

ModelBest and Tsinghua University Open Source Edge-Side Large Model BitCPM-CANN

ModelBest, together with Tsinghua University and the OpenBMB open-source community, has released BitCPM-CANN, China's first 1.58-bit ternary large model trained end-to-end entirely on Huawei's Ascend domestic computing platform. Available in four sizes ranging from 0.5B to 8B, the model uses quantization-aware training to free up approximately six times the video memory during inference while retaining 90% to 97.2% of its capabilities. This release aims to provide critical infrastructure for edg

88Z AI editorial1 min read
Flash

SenseTime Launches Raccoon Desktop Agent

SenseTime has officially launched the Raccoon Desktop Agent, positioning it as an AI office partner that evolves from an advisor to an executor. The product features six core capabilities, including direct reading of over twenty local file formats, browser automation, global shortcut activation, office collaboration integration, local memory learning, and automated scheduled tasks, aiming to enhance office efficiency.

88Z AI editorial1 min read
Flash

Tencent Fully Opens ima copilot

Tencent ima copilot has announced full open access, eliminating wait times. The tool features memory and full-scenario awareness, supporting knowledge base access and integration with major model APIs. The Knowledge Square has upgraded to a capability platform, adding Skill publishing functions and launching official skills like WeChat Read.

88Z AI editorial1 min read
Flash

Zhipu Qingyan Launches AgentMore Skills Square

Zhipu Qingyan has officially launched the AgentMore Skills Square, integrating three major modules: official recommendations, a vast library of free skills, and an open-source community. It supports global search and one-click installation with zero tokens. Users can quickly equip agents with exclusive skill packs such as WeChat Reading and Meituan coupons, enabling instant upgrades from general chat to industry expertise, with flexible uninstallation and management options.

88Z AI editorial1 min read
Flash

Zhipu Launches GLM-5.1 High-Speed API, GLM-5.1-highspeed

Zhipu Open Platform has released the GLM-5.1 High-Speed API, achieving an output speed of 400 tokens/s and setting a new global benchmark for large model API speed. Jointly launched by Zhipu and the TileRT team, it combines flagship capabilities with ultra-low latency. It is suitable for high-demand scenarios like AI coding and real-time interaction, and is now available to select enterprise clients.

88Z AI editorial1 min read