Welcome to AgiYes‘s AI News Weekly!
Don’t have time to keep up with the latest in AI? We’ve done the heavy lifting. Here are the 10 most significant AI developments from this week, simplified for you.
Top 10 AI News from September 6–12, 2025
1. Anthropic Launches New Claude AI Feature
Anthropic has introduced a new feature for its Claude AI chatbot, enabling automatic memory of chat history for team and enterprise users. This enhancement allows Claude to recall previous conversations without manual prompts, improving its understanding of user needs and preferences. Currently, the feature is exclusively available for team and enterprise subscribers
2. Replit Releases Agent 3: A More Autonomous Coding Assistant
Replit has launched Agent 3, its latest AI-powered programming assistant with significantly enhanced autonomy. The new version boasts major improvements in code generation, debugging, and project management, reportedly offering 10 times greater independence than its predecessor. Replit plans to further upgrade Agent 3 with support for more programming languages, better team collaboration, and deeper cloud integration.
3. Stability AI Releases Stable Audio 2.5 for Professional Sound Production
Stability AI has launched Stable Audio 2.5, a new audio generation model designed to help creative teams efficiently produce high-quality, customizable audio content. The model supports multi-segment music creation, including intros, developments, and endings, addressing the growing demand for diverse audio materials in the market.
4. Sarvam Introduces Samvaad Voice and Chat AI Agent on WhatsApp
Sarvam has deployed its Samvaad voice and chat AI agent on WhatsApp, allowing businesses to integrate AI-driven communication directly into the platform. Users can seamlessly switch between voice and text conversations within the same chat, thanks to Meta’s new business call API. The agent supports 11 Indian languages and offers voice, text, and interactive widget-based interactions.
5. Freepik Integrates Seedream 4.0 for Unlimited AI Image Generation
Freepik has integrated ByteDance’s Seedream 4.0 image generation model into its creative toolkit, providing designers with advanced multimodal AI capabilities. Premium+ and Pro users now enjoy unlimited image generation without quotas. The model delivers industry-leading speed and output quality, marking a major upgrade for the platform.
6. Claude Now Generates Excel, PPT, and PDF Files in One Click
Claude AI has introduced a new feature that enables users to generate and edit Excel spreadsheets, Word documents, PowerPoint presentations, and PDF files directly within the platform. The function is currently available to Max, Team, and Enterprise users on Claude.ai and desktop apps, with Pro users gaining access in the coming weeks.
7. Google Expands AI Mode to 5 More Languages
Google has extended its AI mode to support Hindi, Indonesian, Japanese, Korean, and Brazilian Portuguese. This follows last month’s expansion to 180 new markets for English queries. The move signals Google’s effort to transition AI search technology from experimentation to mainstream global use.
8. Microsoft Invests in Training Its Own AI Models and Partners with Anthropic
Microsoft is developing its own computing clusters and training its MAI-1-preview model using 15,000 H100 GPUs—a relatively small-scale setup so far. The company also plans to integrate Anthropic’s AI models into Microsoft 365 Copilot, as these reportedly outperform OpenAI in Excel and PowerPoint tasks, reflecting Microsoft’s diversified AI strategy.
9. Perplexity Raises $200M at $20B Valuation Amid Google Challenge
AI search startup Perplexity has secured $200 million in new funding, reaching a $20 billion valuation—up from $18 billion just two months ago. The three-year-old company has raised $1.5 billion to date and continues to challenge Google’s dominance. Last August, Perplexity proposed acquiring Google’s Chrome for $345 billion amid antitrust allegations, though a recent federal ruling allowed Google to retain its search business.
10. Moonshot AI Open-Sources “Checkpoint Engine” for Efficient LLM Training
Moonshot AI has open-sourced Checkpoint Engine, a middleware tool designed for LLM inference that enables efficient hot-swapping of model weights. It can synchronize a 1-trillion-parameter model like Kimi-K2 across thousands of GPUs in about 20 seconds, significantly reducing downtime during reinforcement learning training.
Final Thoughts on This Week’s AI News
This week, AI advancements focused on integration and efficiency.
Anthropic and Claude are now smarter, offering features like chat memory and direct file creation to streamline workflows.
Replit and Freepik are empowering creators with more autonomous coding and unlimited image generation.
Meanwhile, global access is expanding, with Google adding new languages to its AI search and Sarvam bringing AI agents to WhatsApp.
The market remains highly competitive, as Microsoft diversifies its strategy and Perplexity secures massive funding. Overall, AI is becoming a more practical and essential tool, with continued innovation across the board.
Read More:
- AI News from August 30–September 5, 2025
- AI News from August 23–29, 2025
- AI News from August 16–22, 2025
- AI News from August 9–15, 2025
