Gemini 3.1 Pro Preview Custom Tools is a specialized variant of Gemini 3.1 Pro that targets one specific weakness in agent-style coding workflows: the tendency of the base model to fall back on a generic bash tool when more efficient third-party or user-defined functions are available. The preview endpoint re-trains or otherwise steers that tool-selection behavior so the model picks the most appropriate callable in complex, multi-tool setups, which both excerpts describe as significantly increasing function-calling reliability. It carries forward the core multimodal reasoning strengths of Gemini 3.1 Pro, handling text, image, video, audio, and code within a 1M-token that quick-info value window, and is positioned for production-grade software engineering work where agent orchestration matters more than raw chat quality.
For practitioners, the variant is best understood as a drop-in Gemini 3.1 Pro replacement when reliable, discriminative tool use is the priority, such as in coding agents that must choose between custom APIs rather than shelling out commands. The benchmark and capability signals reported on LM Market Cap reinforce that focus: an overall composite score of 92 with the capabilities sub-score at 100, an LMSYS Arena Elo of 1494, and a Vision Elo of 1278, reflecting strong general reasoning plus solid vision grounding that complements multi-tool pipelines. Within the coding category it ranks near the top tier of tracked models, and its 65,536-token output ceiling gives it room to emit full files, multi-file diffs, or detailed planning traces in a single response, making it well suited to long-horizon agent loops rather than short conversational exchanges.
