Mistral AI launched a document understanding API that processes PDF files using OCR to generate Markdown documents, offering superior performance for mathematical equations, scanned documents, and multilingual content while processing up to 2000 pages per minute.
Google has launched an AI mode for Search: an experimental product that leverages AI to handle complex queries typically addressed by performing multiple traditional web searches. In parallel, the company also announced it expanded access to its AI Overviews, now powered by Gemini 2.0.
Scrunch AI recently exited beta with $4M in seed funding from Mayfield to help businesses maintain visibility in AI-generated search results. Scrunch AI has seen promising adoption, having already secured 25 enterprise customers, including Lenovo and Crunchbase.
Cohere for AI released the Aya Vision models (8B/32B), which support 23 languages, perform remarkably well in image captioning, visual Q&A, and translations, and outperform competitors in benchmarks. Aya Vision is available via Cohere Playground, WhatsApp, Kaggle, and Hugging Face.
Anthropic announced earlier this week that it has secured $3.5B in a Series E funding round led by Lightspeed Venture Partners, with new and existing investors participating. This Series E announcement follows the Claude 3.7 Sonnet and Claude Code launches.
OpenAI has released GPT-4.5, its largest AI model to date, featuring improved knowledge and emotional intelligence. But while GPT-4.5 is a clear improvement over GPT-4o, it falls behind the o series in benchmark evaluations and comes with dramatically higher costs.
Anthropic launches Claude 3.7 Sonnet, the first hybrid reasoning AI that offers both quick responses and visible step-by-step thinking. It excels at coding tasks and comes with Claude Code, a new terminal tool for developers.
xAI launched Grok 3 and Grok 3 mini with reasoning capabilities and a DeepSearch agent. Shortly after their release, the models sparked controversy when xAI compared their consensus-based math scores to competitors' single-attempt metrics, highlighting ongoing issues with benchmark comparisons.
Meta announced LlamaCon, its first developer conference for its generative AI ecosystem on April 29. The LlamaCon announcement follows a year of rapid Llama model releases that gained 650M+ downloads and 85,000+ derivatives.
Mistral AI has launched Mistral Saba, a language model optimized for Arabic that works well with Indian-origin languages, positioning the startup to expand its presence in Middle Eastern and South Asian markets.
Lingo.dev, a startup offering AI-powered UI localization automation for developers, has secured $4.2 million in seed funding to provide comprehensive app, database, and website translation services beyond simple text conversion.
Former OpenAI CTO Mira Murati has launched Thinking Machines Lab, a new AI startup focused on making artificial intelligence more customizable and accessible while advancing AI capabilities, with a team of prominent AI researchers from OpenAI and other leading companies.