Brief

ZeroGPU has launched a new Batch API designed to efficiently tag large volumes of unstructured text data, such as customer reviews, at a significantly lower cost than traditional synchronous API calls. The service utilizes a specialized, smaller model called LFM2.5-1.2B-Instruct, which is optimized for tasks like sentiment analysis and topic classification, avoiding the expense of using larger, more complex frontier models for repetitive work. A new cookbook demonstrates how to use the Batch API for overnight processing of CSV files, with features for smart error handling and seamless merging of results back into existing databases. AI

The llmfleet library introduces a novel approach to optimizing API calls for large language models, particularly Anthropic's Batch API. It addresses the limitations of the current API design by pooling multiple agent requests into a single batch, potentially saving 50% on input token costs. The library's dispatcher intelligently routes requests based on a specified latency budget, allowing for both fast, synchronous responses and slower, batched processing without the caller needing to manage the complexity. AI