Manus AI agent can do 'anything'

Welcome back! Here’s what you need to know about AI today:

👉 Chinese startup released an AI agent that can perform any task

👉 New open-source math model is trained with $1K and is better than DeepSeek

👉 Google is adding Gemini to its calendar

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 5 minutes

What’s going on: Manus, an 'agentic' AI platform by Monica, a Chinese startup, has generated significant hype, with some calling it the most impressive AI tool ever and comparing it to a breakthrough on par with DeepSeek. Manus is designed to bridge the gap between thought and execution, allowing it to perform complex tasks without requiring constant human input. Unlike traditional AI chatbots that need user instructions to function, Manus can autonomously gather information and make decisions, such as finding an apartment based on various factors like crime rates and weather conditions.

What does it mean: The architecture of Manus AI is built around multiple sub-agents that specialize in different areas, enabling it to handle intricate multi-step processes efficiently. This asynchronous operation allows Manus to work in the background, notifying users only when tasks are completed. Its capabilities span a wide range of applications, including data analysis, content generation, and professional report writing. The platform has reportedly achieved over 65% accuracy on the GAIA benchmark, which evaluates AI systems on their reasoning and task completion abilities.

More details: 

• Manus uses a combination of existing and fine-tuned AI models, including Anthropic's Claude and Alibaba's Qwen, instead of being developed entirely from scratch.

• Manus has sparked discussions about its impact on the coding industry and job displacement. While it automates routine coding tasks and could enhance productivity for developers, people are concerned regarding potential job losses in sectors relying on basic coding skills.

• User have experienced problems like error messages, endless loops, mistakes on factual questions, inconsistent citations, and crashes.

• Interested? Manus is not officially launched. However, for using the beta version of it you should have an invitation code. Visit their website and join the waitlist to get yours!

What’s going on: Researchers have recently launched Light-R1-32B, an open-source AI model specifically designed for advanced mathematical problem-solving. This model, which boasts 32 billion parameters, has demonstrated superior performance compared to other models, including DeepSeek, on rigorous benchmarks like the AIME. Remarkably, Light-R1-32B was trained in under six hours using 12 Nvidia H800 GPUs at a minimal cost of approximately $1,000. This efficiency positions it as a highly accessible option for organizations seeking to leverage AI for complex mathematical tasks.

What does it mean: Light-R1 creators used a curriculum-based supervised fine-tuning approach combined with direct preference optimization to enhance the model’s mathematical reasoning capabilities. The model's performance metrics are impressive; it achieved a score of 76.6 on AIME24 and 64.6 on AIME25, surpassing DeepSeek's scores of 72.6 and 61.9 respectively.

More details:

• Light-R1-32B is released under the Apache 2.0 license, which allows for free usage and modification by enterprises and researchers alike.

• The researchers have also made available all training datasets and evaluation tools. For downloading/using model source and datasets, check out their Hugging Face page.

🍎 Apple's smart home hub launch has been delayed because the upgraded Siri features it relies on are taking longer than expected to develop.

⚖ Elon Musk may still have a chance to prevent OpenAI's for-profit conversion, as a federal judge, expressed concerns about the transition and offered an expedited trial in the fall of 2025. Musk recently lost a battle in his lawsuit against OpenAI.

🤖 Microsoft is intensifying its AI efforts to compete with OpenAI, by developing its own AI models like MAI and exploring alternative models from companies like xAI and Meta for use in products like Copilot.

👩‍⚖️ A federal judge has allowed a copyright lawsuit against Meta to proceed, with authors alleging Meta used their books to train its Llama AI models and removed copyright information.

💰 Cursor, famous AI code editor, is reportedly in talks to raise capital at a valuation of nearly $10 billion, just three months after a previous fundraise that valued the company at $2.5 billion.

🦙 Meta is planning to introduce improved voice features, including the ability for users to interrupt the model mid-speech, with Llama 4, its next flagship AI model expected in the coming weeks.

🏭 xAI, has acquired a 1 million-square-foot property in Memphis to expand its AI data center, aiming to upgrade its primary facility to 1 million Nvidia GPUs this year.

📅 Google is adding a Gemini-powered side panel to Google Calendar, available through Google Workspace Labs, allowing users to manage their schedule conversationally by checking details, creating events, and more via prompts.

AI + Real estate agent

VISA - Senior Machine Learning Engineer - Cloud AI Platform

Latitude - Data Services Analyst II, Mapping (Contractor)

Docugami - Machine Learning / Data Scientist

Curaleaf - Data Scientist

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!