Gemini 3 Pro

Gemini 3 Pro is a multimodal language model from Google DeepMind, released on February 19, 2026. It scores 90.1 on MMLU-Pro, 80.6 on SWE-Bench Verified, and holds an Arena ELO of 1485, positioning it among the leading 2026 models for reasoning, code generation, and instruction-following. The model accepts text, code, images, and video inputs, with API pricing set at $2.00 per million input tokens and $12.00 per million output tokens. Its significance lies in establishing a new performance baseline for the Gemini family: the 90.1 MMLU-Pro result represents a 3.2-point gain over Gemini 2.5 Pro’s 86.9, while the 80.6 SWE-Bench Verified score improves on its predecessor’s 71.2 by 9.4 percentage points, demonstrating concrete advances in both knowledge benchmarks and real-world software engineering tasks. However, the shared alias structure with earlier Gemini versions complicates direct predecessor comparisons, making it essential for developers to verify family lineage and benchmark provenance when evaluating the model against alternatives.

Two-hop subgraph: this entity, every entity it directly relates to, and every entity those neighbors relate to. Drag a node, scroll to zoom, click to inspect — or click any neighbor and re-center the atlas there.