Butterfly Effect Company Unveils Manus 2.0 with Cascade Agent and Cue Application for Personal Assistants
Read more
Pandaily
pandaily.com

Butterfly Effect Company Unveils Manus 2.0 with Cascade Agent and Cue Application for Personal Assistants

Butterfly Effect, the developer of the AI agent Manus, released Manus 2.0 to users outside of China on September 28th. Simultaneously, Cue—a new autonomous application for personal agents—was introduced. The company's official WeChat account announced the formation of a team to develop a product for the Chinese market, as well as stable advancement in cooperation with local model developers and ecosystem partners, although the release date for the internal version has not been specified.

The company describes Manus 2.0 not as another update, but as an entirely new architecture. At the core of this architecture is Cascade—the latest version of the internal Manus agent framework. Cascade is designed to support project lightweightness from the outset, connecting specialized capabilities only when required to perform a task. According to Manus tests, in one configuration, Cascade allowed for 23.2% fewer tokens, completed tasks 28.2% faster, and reduced launch costs by 32% compared to the previous system.

The architecture is accompanied by two key infrastructure features. Users can now purchase a Cloud Computer—a dedicated environment for projects such as a multiplayer game server or automation that must run continuously even when the user's laptop is closed. Automations have expanded beyond simple scheduled tasks and can now be triggered by event triggers: receiving a new email, changes in ad metrics, a calendar event, a Slack message, or an update in Notion can initiate a workflow configured with just one command.

The desktop application transforms into Manus Studio—a collaborative workspace that loads only the tools necessary for the project, covering documents, spreadsheets, PDFs, presentations, websites, code, games, and videos. The new Video Editor creates a draft version of short videos, such as product commercials, after which it provides the user with an editable timeline featuring separate clips, images, text, graphics, and audio. The Game Dev function combines video, image, and coding models, allowing games to be published as websites or made multiplayer via Cloud Computer. Thanks to Remote Control and Computer Usage capabilities, Manus can operate on the user's own computer in an authorized session, using only approved files, browsers, and applications.

The Cue application functions on both mobile and desktop, utilizing the same infrastructure. Each agent receives its own email address, phone number, wallet, and computer, enabling it to send messages, make payments within a set budget, answer calls, and compile summaries. Multiple agents can participate in group chats and pass work to each other; an agent can also order food at a restaurant after the user scans a QR code. Cue is available in a free, invitation-only early access mode, while the iOS version awaits review in the App Store. Founder Xiao Hong shared on the social network Jike his view that something possessing its own phone number, email, payment methods, computer, and sufficient intelligence can be called a personality.

Similar stories

China Telecom releases Xing4.0-29B-A4B model trained on Ascend with MoE architecture
Read more
pandaily.com

China Telecom releases Xing4.0-29B-A4B model trained on Ascend with MoE architecture

China Telecom Artificial Intelligence Technology Co., Ltd. has made the Xing4.0-29B-A4B model available, which is the latest development in the Xing series (formerly known as TeleChat). The weights and configuration files for this model are hosted on Hugging Face and ModelScope under the XingChen-AGI organization. Officially, the model is positioned as China Telecom / XingChen / Xing4.0.

The model is presented as an agentic implementation of Mixture-of-Experts (MoE), optimized for the Ascend platform, rather than just a general release of a large language model. According to company materials, this is the first model of this scale fully trained on Ascend NPUs using the MindSpore framework and specifically tuned for complex engineering tasks and agent workloads.

According to Hugging Face data, the model has a total of 29 billion parameters, with approximately 4 billion parameters activated per token. It supports a native context length of 256K, which can be extended to 512K, features 40 layers, uses MLA attention, and includes 64 routed experts, of which four experts plus one shared expert are activated per token. Architectural notes point to an mHC + MLA + MTP stack designed for multi-step planning, tool calling, and maintaining coherence over long contexts.

It is reported that joint optimization of training on Ascend 910C clusters via MindSpore/MindFormers, including integrated mHC operators and MoE communication tuning, increased training throughput by approximately 96% compared to baseline settings, according to vendor data.

Various serving paths are provided for usage, such as Transformers, vLLM, SGLang, and KTransformers. Fine-tuning support is also available through LLaMA-Factory and MindFormers, along with formatting for agent systems, including OpenCode, Claude Code, OpenClaw, and Hermes. Furthermore, Phoenix Tech notes that coverage of secondary IT topics shows agent results comparable to leading Qwen models, and claims that 4-bit quantization can reduce memory consumption to 15GB for local long-context inference; however, these statements remain third-party or company claims pending independent verification.

For teams working on Huawei Ascend stacks, the Xing4.0-29B-A4B weights are provided with explicit training provenance on MindSpore and open interfaces for inference within the Apache ecosystem. Thus, this is an OSS agentic MoE from China Telecom, not a closed announcement based solely on an API.

Popular