AI labs shift toward task-specific models

OpenAI, Anthropic and Google DeepMind are shifting from broad general-purpose AI systems toward models designed for specific workloads. The newer lineups, including Sol, Sonnet 5 and Gemini 3.1, are positioned around faster responses, stronger task execution and more cost-efficient performance for businesses and developers.

OpenAI’s GPT-5.6 lineup includes Sol, Terra and Luna, each aimed at different needs. Sol focuses on complex tasks and safety features, Terra is described as powerful while costing half as much, and Luna is built for fast, low-cost applications. GPT-5.6 also adds developer controls for setting task effort and an autonomous helper mode.

Anthropic’s Claude Sonnet 5 adds default thinking capabilities for planning, testing and executing tasks, including codebase debugging. Its tokenizer is said to generate about 30% tokens, helping with outputs such as research briefs and spreadsheets. Google DeepMind’s Gemini 3.1 and 3.5 Flash add a 3-tier thinking system, while Gemini 3.1 Pro supports outputs up to 65,536 tokens and can generate animated SVGs and complex 3D code structures from text prompts.