Gemini 2.5 Pro Experimental 0325

Gemini 2.5 Pro Exp 0325. Google's experimental model from March 2025.

Reasoning
Vision
Audio Input

Added Mar 25, 2025

Pricing

Auto routing · per 1M tokens

Input: $2.50
Output: $10.00
Cache read: $0.25

Compare provider prices

Specifications

Context window: 1M
Max output: 65.5K

Benchmarks

Artificial Analysis

LMArena

Vectara

Sourced from Artificial Analysis.

Intelligence Index

16.1

Better than 58% of models compared

Coding Index

33.3

Better than 40% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

2.2%

Better than 21% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

298 Elo

Better than 16% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

459 Elo

Better than 24% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

10.2%

Better than 41% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

69.0%

Better than 60% of models compared

Reasoning

HLE

Humanity's Last Exam

22.5%

Better than 71% of models compared

IFBench

Instruction-following benchmark

48.7%

Better than 58% of models compared

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

0.0%

Better than 16% of models compared

SciCode

Python programming for scientific computing

46.3%

Better than 43% of models compared

LiveCodeBench

Contamination-free coding benchmark

80.1%

Better than 92% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

87.7%

Better than 85% of models compared

AIME

American Invitational Mathematics Examination

88.7%

Better than 95% of models compared

Math-500

Diverse mathematical problem solving benchmark

96.7%

Better than 86% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

86.2%

Better than 95% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

84.4%

Better than 76% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

26.5%

Better than 68% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

54.1%

Better than 55% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation