Gemini 3 Deep Think (source unspecified)
This benchmark entry preserves generative-model observations labeled Gemini 3 Deep Think when the evaluator does not identify whether it tested the original 2025 release or the February 2026 upgrade. A generic display label or restricted evaluation endpoint alone does not establish the underlying snapshot. Exact February 2026 observations and the explicitly dated MathArena December 2025 evaluation remain separate.
Benchmark scores
| Benchmark | Score |
|---|---|
| LiveCodeBench Pro | 3298 |
| VoxelBench Text Rating | 1468 |