Gemini 3.1 Pro Preview AI Evaluation | LayerLens - Benchmark Results & Performance

Model

Gemini 3.1 Pro Preview by Google

text+image->text

Free credits included. No credit card required.

Gemini 3.1 Pro Preview is Google’s frontier reasoning model, delivering enhanced software engineering performance, improved agentic reliability, and more efficient token usage across complex workflows. Building on the multimodal foundation of the Gemini 3 series, it combines high-precision reasoning across text, image, video, audio, and code with a 1M-token context window.

Evaluations

across 16 benchmarks

Latency

21.4s

Context length

tokens

Cost

$2 · $12

input · output per 1M tokens

Key takeaways

Google's Gemini 3.1 Pro Preview is a multimodal reasoning model designed for complex workflows in software engineering, aiming to enhance agentic reliability and token efficiency.
The model utilizes a MoE (Mixture of Experts) architecture and features a large 1M-token context window.
Gemini 3.1 Pro Preview demonstrates exceptional capabilities in advanced mathematical problem-solving, achieving 100% accuracy on the AIME 2024 Part I benchmark, consistently providing clear, step-by-step solutions for arithmetic, algebraic, and geometric tasks.
Despite its strengths, the model struggles significantly with complex multi-step system configurations, nuanced security protocols, network interactions, and specific data processing tasks, as evidenced by a low 38.5% pass rate on the 'Terminal-Bench (Terminus-1)'.
Key limitations include inconsistent LaTeX output, occasional misinterpretations of mathematical problem constraints, and difficulty in orchestrating multiple system-level tools and APIs for real-world development and operational tasks.

Technical specifications

Context length: 1M tokens
Max output: —
Architecture: MoE
Modalities: text+image->text
Parameters: —
Open weights: No
Licence: Proprietary
Regions: USA
Vendor: Google
Provider model key: google/gemini-3.1-pro-preview