Gemini 3.1 Pro
Google DeepMind · February 2026
● activeClosedsparse moemultimodalAPI Available
Parameters~1T (MoE)
Context Window2M tokens
Why It Matters
Pushed the context window to 2M tokens and more than doubled reasoning benchmark scores, signaling Google's commitment to both scale and intelligence improvements.
Description
Google's frontier reasoning model with roughly 1 trillion total parameters using sparse mixture-of-experts (only a fraction of the model activates per query for efficiency). Features a 2-million-token context window — enough to process several books or massive codebases simultaneously. More than doubled reasoning performance compared to Gemini 3 Pro on the ARC-AGI-2 benchmark.
Notable Milestones
- ▸Scored 77.1% on ARC-AGI-2 reasoning benchmark
- ▸2M token context enables processing of entire repositories
Key Innovations
Long Context
Long ContextAbility to process very long inputs (100K+ tokens), enabling analysis of entire codebases or books.
MoE
MoEArchitecture where only a fraction of the model's parameters are active for each input, allowing massive scale with lower compute.
Reasoning
ReasoningStructured step-by-step problem solving, often using chain-of-thought or tree-of-thought approaches.
Family Tree
Built On
Lineage
Successors (2)
Enabled By
TPU v6 (Trillium)GOOGLE · June 2025
Higher BF16/FP8 FLOPS