Google Announces Veo 2 Pricing at $0.50 per Second; Kimi.ai and Ozone AI Release New Models
New AI Models
Google Releases Veo 2: Google has launched Veo 2, an AI model that generates high-quality videos from text prompts at a cost of $0.50 per second. The model is praised for its consistency, quality, and prompt understanding but has raised concerns about its censorship policies and lack of certain features.
Kimi.ai Releases Moonlight: Kimi.ai has released Moonlight, a 3B/16B Mixture of Experts (MoE) model trained with their improved Muon optimizer. The model outperforms similar state-of-the-art models in most benchmarks and is notable for its speed and efficiency, particularly in CPU inference.
Ozone AI Releases Chirp-3b: Ozone AI has introduced Chirp-3b, a new AI model trained on over 50 million tokens of distilled data from GPT-4o. The model has shown impressive performance on benchmarks like MMLU Pro and IFEval.
AI Benchmarking and Performance
DeepSeek's Long Context Comprehension: A benchmark comparison of various AI models' performance in handling long context comprehension shows that models like o1 and deepseek-r1 perform well up to certain context lengths, but others struggle with longer contexts.
Coding Performance Benchmarks: Grok-3 scored significantly lower than o3-mini-high in coding tasks, highlighting the varying performance of different AI models.
New AI Products and Services
Chrome Extension for Image Filenames: A new Chrome extension uses local AI models (LLaVa and Ollama) to analyze image content and generate descriptive filenames automatically. The extension runs entirely on the user's machine, ensuring privacy and speed.
Live Interpreter App: A new iOS app uses the gpt-4o-mini model for live transcription and translation, effectively translating idioms. The app also utilizes the Whisper speech-to-text model on Groq Cloud at a lower cost.