Scale AI's AgentEnv framework now supports running evaluations on Vercel Sandbox, allowing for task scoring and action recording. Available from version 0.9.1334, it uses separate Sandboxes for agents and tasks.
Related articles
See all- ChromaDB
ChromaDB: Embeddings and Vector Databases in Python
Real Python introduces embeddings and vector databases with ChromaDB, explaining how relevant documents can provide context for LLMs. The supplied excerpt covers vector fundamentals and NumPy operations; the tutorial targets ChromaDB 1.5 and Python 3.12+.
- OpenAI
GPT-6.1 Terra now available on AI Gateway
OpenAI's GPT-6.1 Terra, a fast model for visual reasoning and text processing, is now available on AI Gateway. It is suitable for high-volume tasks in desktop applications, though larger GPT-6 models are recommended for complex coding tasks.
- OpenAI on OpenAI
How Ultrafast cut incident investigation from hours to minutes at OpenAI
OpenAI engineers have reduced incident investigation time from hours to minutes using Codex Ultrafast, enhancing security analysis, code refactoring, and daily workflows.
- Mobile
Google ML Drift Accelerates On-Device AI Inference Across Diverse GPUs
Google has introduced ML Drift, an open-source GPU compute engine for on-device AI and machine learning inference. It simplifies the complexities of different GPU hardware and low-level APIs, supporting OpenGL ES, OpenCL, Metal, and WebGPU.
- Image Storage
Three Ways of Storing and Accessing Lots of Images in Python
Real Python introduces PNG files, LMDB, and HDF5 as image-storage options. The available excerpt explains the motivation and CIFAR-10 setup, but does not include performance or disk-usage comparison results.
- AI
Introducing Kimi K3.1 Pro
Kimi K3.1 Pro has been introduced on the Kimi API, with full model weights expected by November 4, 2026. It offers a "Highspeed" version that's 2-3x faster at double the price, and competes closely with leading models like Claude Opus 5.5 and GPT-6 Astra in benchmarks.