Google has updated its flagship Gemini model to version 3.1 Pro, highlighting improved reasoning as measured by a new benchmark.

Benchmark results

In an announcement the company emphasized the ARC-AGI-2 test, which evaluates a model’s ability to solve novel logical patterns under uncertainty.

On the ARC-AGI-2 assessment Gemini 3.1 Pro achieved a score of 77.1%, more than double the previous Gemini 3 Pro result.

Google states the upgrade represents progress in foundational reasoning and shows stronger performance on complex problem-solving tasks across multiple domains.

Performance in other benchmarks

The company also reported competitive outcomes in code generation benchmarks and in evaluations involving autonomous AI agents and related workloads.

Google did not provide detailed leaderboard placements for all tasks in the announcement but indicated improvement relative to the previous release.

Availability

Gemini 3.1 Pro is available now through Google’s AI Studio platform for developers and enterprise users accessing model endpoints and tools.

Google recommended users review documentation and resource guidance in AI Studio to understand deployment constraints and cost implications before production use.

Independent verification and third-party evaluations will be needed to assess how the reported gains translate to real-world applications and end-user outcomes.