What happened
Google released Gemini 3.1 Pro in preview, scoring 77.1% on ARC-AGI-2 — more than double the reasoning performance of Gemini 3 Pro. The model ships with a 1M token context window and is with competitive pricing across its Flex and Priority tiers, positioning it as a strong price-to-performance option among closed frontier models. Available through the Gemini API, AI Studio, Vertex AI, Gemini CLI, and Android Studio.
Why it matters
ARC-AGI-2 tests novel reasoning — problems the model has never seen before, not pattern matching from training data. Doubling this score in a single generation suggests Google found a genuine reasoning breakthrough, not just scale. With competitive pricing and 1M context, Gemini 3.1 Pro is positioned as the default choice for cost-sensitive agentic workloads that still need strong reasoning. The quiet part: Google gained the most ground in the model race this quarter while Anthropic and OpenAI dominated headlines.
Who should pay attention
Developers building cost-sensitive AI applications, teams evaluating models for complex reasoning tasks, enterprise architects comparing frontier model pricing, and anyone building agents that need to solve novel problems rather than retrieve memorized patterns.