Google: Gemini Pro Latest

Gemini Pro Latest is a rolling API alias (gemini-pro-latest) that always resolves to the newest model in Google's Gemini Pro family, so capabilities and benchmark scores track Google's current flagship Pro release rather than a fixed checkpoint. As of early 2026 the alias routes to Gemini 3.1 Pro Preview, a closed multimodal reasoning model served through the Gemini API, Google AI Studio, Vertex AI, and the Gemini app.

Gemini 3.1 Pro is built for deep reasoning, long-context work, agentic tool use, and advanced coding. It accepts text, images, audio, video, and PDFs, supports up to about one million input tokens and large text outputs, and exposes thinking-depth controls for harder tasks. Google reports strong gains over Gemini 3 Pro on graduate-level reasoning (Humanity's Last Exam), scientific QA (GPQA Diamond), agentic coding (SWE-bench Verified, Terminal-Bench 2.0), competitive programming (LiveCodeBench Pro Elo), multimodal understanding (MMMU-Pro), web research (BrowseComp), and long-context recall (MRCR v2 at 128k).

Because the alias hot-swaps when Google ships a newer Pro model, production systems should pin an explicit model ID (for example gemini-3.1-pro-preview) if they need reproducible behavior, latency, or pricing. For exploratory use and access to the latest Pro-tier quality without manual version updates, Gemini Pro Latest is the intended entry point.