The models keep coming. French AI startup Mistral has launched two new large language models (LLMs) based on the Mamba architecture: Codestral Mamba and Mathstral. Codestral Mamba offers faster code generation with longer input texts, while Mathstral is designed for math-related reasoning and scientific discovery. Mistral’s models outperformed rival open-source models in benchmarking tests and can be accessed through Mistral’s la Plateforme and HuggingFace.
OpenAI has announced the release of GPT-4o mini, a smaller and more cost-efficient version of its AI model. Priced at USD 0.15 per 1 million tokens for input and $0.60 for output, GPT-4o mini aims to reduce costs for developers using OpenAI’s APIs. The model outperforms GPT-3.5 Turbo on various benchmarks and can handle text and vision inputs. It is also expected to be available on Apple devices through Apple Intelligence. While GPT-4o mini replaces GPT-3.5 Turbo in ChatGPT, the older model will still be supported to avoid breaking existing applications.
Moving beyond models, Microsoft has introduced a new AI system called SpreadsheetLLM, designed to understand and work with spreadsheets. The system combines large language models with structured data found in spreadsheets, enabling AI to reason over spreadsheet contents. This opens up possibilities for automating data analysis tasks, providing intelligent insights, and democratizing access to data insights.
AWS has introduced AWS App Studio, a generative AI-powered service that allows users to create enterprise-grade applications in minutes without requiring software development skills. The service uses natural language to understand app descriptions and provides template prompts to refine ideas. Additionally, AWS has enhanced its agents for Amazon Bedrock, enabling them to remember previous interactions and run code snippets.
Why do we care?
I’m intrigued by more specialized models. Solution providers are all about matching technology to problem, and it’s worth knowing the variety of models out there. One can see a future of a single, general AI assistant that leverages specialized models when needed.