Full Summary
This Thursday morning, multiple sources including iNews Zoombangla and CNET confirm China's Kimi K3 AI model, with 2.8 trillion parameters, overwhelmed its servers in under 72 hours, forcing Moonshot AI to halt new subscriptions. Kimi K3 is the largest open-weight AI released to date, matching or outperforming models like Claude Opus and GPT-4 Turbo. This comes as both OpenAI and Alibaba launched new voice AI models for developers. Dev.ua reports OpenAI expanded its API with GPT-Live-Transcribe for real-time tasks and GPT-Transcribe for pre-recorded audio, significantly improving semantic accuracy. Simultaneously, Alibaba introduced Qwen-Audio-3.0-Realtime Plus and Flash models for direct speech-to-speech interaction. Meanwhile, India Today and Times Now report that Indian startup Sarvam AI is building a foundational AI model with over a trillion parameters, aiming to challenge OpenAI and Anthropic. This model, confirmed by Whalesbook to be released within six months, will focus on coding, cybersecurity, and simulation, offering significantly lower costs than competitors. In other developments, Memeburn states Amazon is winding down several Nova AI models, shifting focus to a new Frontier Model Research team led by Pieter Abbeel. Tech-insider.org reveals Meta's new Muse Spark 1.1 is up to 83% cheaper than Anthropic's Claude Opus 4.8, marking Meta's entry into selling model access via a proprietary API. Thewincentral.com reports rumors of an interim GPT-5.7 launch from OpenAI in August, with GPT-6 potentially pushed to September. This rapid global AI expansion, particularly with open-source and cheaper alternatives, means consumers and businesses can expect more powerful and accessible AI tools, but also face increasing questions about AI safety and the geopolitical landscape of technology.