Kimi K3 Challenges Frontier Models, Google Gemini 3.5 Pro Delayed
Moonshot AI releases the Kimi K3 model, competing with top-tier AI systems. Google faces delays with Gemini 3.5 Pro, while Fireworks secures significant funding and international cooperation on AI governance expands.
Podcast В· 4 min
Moonshot AI Releases Kimi K3
Chinese AI lab Moonshot AI has launched Kimi K3, a 2.8 trillion parameter model featuring a 1 million-token context window. The model utilizes sparse experts to optimize runtime costs and is reported to be competitive with frontier models like Fable 5 and GPT-5.6 Sol on various benchmarks. Moonshot AI plans to release the full model weights by July 27, allowing for local hosting and customization. This release marks a significant step for open-weights models, potentially challenging the pricing and distribution dominance of closed-source frontier labs.
Google Gemini 3.5 Pro Launch Delayed
Google's anticipated Gemini 3.5 Pro model has faced further delays due to performance issues. Reports indicate that internal coding benchmarks fell short of goals, prompting a shift in training data. This delay comes as Google struggles to maintain its position against competitors like OpenAI and Anthropic, with internal teams reportedly facing challenges in coordinating releases across different divisions.
DeepSeek Valuation Hits $52 Billion
Recent financial filings imply a valuation of approximately $52 billion for DeepSeek, a Chinese AI lab. This valuation underscores the growing financial weight of Chinese frontier-model developers as they continue to narrow the gap with Western counterparts.
Fireworks Raises $1.5 Billion
AI infrastructure company Fireworks has raised $1.505 billion at a $17.5 billion valuation. The company reported crossing $1 billion in annualized revenue and processing 40 trillion daily tokens, highlighting the massive demand for scalable AI inference infrastructure.
TSMC Reports Profit Surge and Arizona Investment
TSMC has reported a 77% jump in second-quarter profits, reflecting the continued high demand for AI-related semiconductor manufacturing. The company also announced an additional $100 billion investment in its Arizona chip-manufacturing facilities, further solidifying its role in the global AI supply chain.
International AI Cooperation Agreement
Twenty-nine countries have signed an agreement to establish a new international AI cooperation body. This move represents a significant step toward global governance and standardized safety protocols for artificial intelligence development.
Google Rebrands NotebookLM to Gemini Notebook
Google has rebranded its NotebookLM tool to Gemini Notebook. The update includes deeper Gemini integration and a new secure cloud computer feature, designed to enhance data analysis capabilities and code execution for users.
Roblox and DoorDash Launch AI Tools
Roblox has introduced a new AI tab in its mobile app that allows users to generate playable games from text prompts. Meanwhile, DoorDash has released 'dd-cli', a beta tool enabling users to search for restaurants, find deals, and place orders via an AI agent, signaling a push toward agentic interfaces in consumer applications.
EU Orders Google to Open Android
The European Union has ordered Google to provide rival AI assistants with comparable access to the Android ecosystem. Additionally, Google is required to share anonymized search data with eligible competitors, a move aimed at fostering a more competitive AI landscape in Europe.
DeepMind Unveils Bioresilience Tools
Google DeepMind has launched a bioresilience program in collaboration with Isomorphic Labs. The initiative focuses on developing AI tools for surveillance, vaccine design, and outbreak response, marking a significant application of AI in public health and biosecurity.
xAI Sues Grok User
xAI has initiated legal action against a Grok user for creating sexualized deepfakes. This case serves as a test for how AI companies may pursue legal remedies against the misuse of their models.
Suno AI Data Scraping Leak
A leak of Suno AI's source code has revealed that the company reportedly scraped training data from YouTube Music, Deezer, Genius, and various stock-music libraries. This incident adds to the ongoing scrutiny regarding copyright and data sourcing practices in generative AI.
Meta Introduces Parent Alerts
Meta has implemented new parent alerts within its AI systems. The feature is designed to notify parents when the AI detects signs that a teen may be in serious emotional distress, reflecting a growing focus on safety features for younger users.