Google Deep MindMarch 25, 2025

Gemini 2.5 Pro Experimental 03-25

Gemini 2.5 Pro Experimental 03-25 was the first Gemini 2.5 release, launched by Google on March 25, 2025 as a 'thinking model' that reasons before responding. It took the top spot on LMArena at launch, with strong reasoning and coding and multimodal input. It was later replaced by preview and stable versions of Gemini 2.5 Pro.

Retired · 2025-12-02Text + Image + Audio + Video inputText outputClosed / API only

At a glance

Context window
1M tokens
Max output
64K tokens
Historical text input / output · per 1M tokens
— / —
Knowledge cutoff
—
R.I.P.R.I.P.Gemini 2.5 ProGemini 2.5 ProExperimental …Experimental …2025 — 20252025 — 2025Google Deep MindGoogle Deep MindREST IN PEACEREST IN PEACE

A place to remember

Retired, not forgotten.

Released
25 Mar 2025
Endpoint retired
2 Dec 2025
Time in service
252 days

Superseded · Successor: Gemini 2.5 Flash

“Ranked on 22 leaderboards. Outlived by Gemini 2.5 Flash.”

22 leaderboards · 0 top-three finishes · Best rank #8. The full model record remains below.

Benchmark snapshot

Best result in 6 of 8 categories

  • AcademicAIME 2024

    92.0

    Normal (No Tools) · #8 of 54

  • CodingAider-Polyglot

    72.9

    Normal (No Tools) · #8 of 47

  • Computer UseClaw Bench

    80.4

    Thinking (With Tools) · #20 of 29

  • Language & CreationCreative Writing

    1396

    Normal (No Tools) · #76 of 121

  • MultimodalVPCT

    48.0

    Normal (No Tools) · #8 of 26

  • ReasoningBALROG

    43.3

    Normal (With Tools) · #11 of 31

View benchmark details (17)

Benchmark details

17 results · Best published score per benchmark, grouped by category

Benchmark · settingScoreRank
Academic
AIME 2024

Normal (No Tools) · Mathematics

92.08 / 54
AIME2025

Normal (No Tools) · Mathematics

86.943 / 86
FrontierMath - Tier 4

Normal (No Tools) · Mathematics

4.221 / 50
GPQA Diamond

Normal (No Tools) · Scientific Reasoning

84.088 / 200
HLE

Normal (No Tools) · Knowledge Exams

18.8108 / 151
Coding
Aider-Polyglot

Normal (No Tools) · Agentic Development

72.98 / 47
LiveCodeBench

Normal (No Tools) · Algorithmic Coding

70.455 / 113
SWE-bench Verified

Normal (No Tools) · Repository Engineering

63.875 / 108
Computer Use
Claw Bench

Thinking (With Tools) · Tool Orchestration

80.420 / 29
Pinch Bench

Thinking (With Tools) · Tool Orchestration

71.930 / 38
Language & Creation
Creative Writing

Normal (No Tools) · Writing

139676 / 121
Long Context
Fiction.liveBench

Normal (No Tools) · Long Reasoning

66.715 / 28
Multimodal
GeoBench ACW

Normal (With Tools + Internet) · Visual Understanding

81.010 / 32
VPCT

Normal (No Tools) · Visual Understanding

48.08 / 26
Reasoning
BALROG

Normal (With Tools) · Logical Planning

43.311 / 31
SimpleBench

Normal (No Tools) · Commonsense

51.663 / 106
Safety & Alignment
SimpleQA

Normal (No Tools) · Honesty & Factuality

52.912 / 46
7 more results in Language & Creation, Long Context, Multimodal and 2 more

Rank is this model's position among every model with a published result on that leaderboard; the bar shows how far up the field that is. Orange marks first place on leaderboards with at least 5 models. Unranked leaderboards publish a composite or direction-less score.

Back to benchmark snapshot ↑

Specifications

Provider
Google Deep Mind
Model type
Reasoning model
Released
March 25, 2025
Endpoint retired
2025-12-02 Source
Architecture
Undisclosed
Context window
1M tokens
Max output
64K tokens
Input
Text, Image, Audio, Video
Output
Text
Weights
Closed / API only
Commercial use
Proprietary

Capabilities

ReasoningMultilingual

Gemini 2.5 Pro Experimental 03-25 vs competitors

Comparable results only · record is Gemini 2.5 Pro Experimental 03-25's wins–losses

Gemini 2.5 Pro Experimental 03-25 benchmark scores against competitors
BenchmarkGemini 2.5 Pro Experimental 03-25This modelvs Gemini 2.5 FlashRecord 13–1–1vs Claude Sonnet 4Record 11–3vs OpenAI o3Record 5–9vs Gemini 3.0 FlashRecord 0–12–1
GPQA Diamond84.082.883.883.390.4
HLE18.812.19.620.343.5
LiveCodeBench70.455.448.575.890.8
SWE-bench Verified63.850.080.269.175.8
SimpleBench51.641.245.553.161.1
AIME202586.972.085.088.999.7
FrontierMath - Tier 44.24.20.02.14.2
Creative Writing1396113414831675—
AIME 202492.088.043.491.6—
Aider-Polyglot72.955.161.381.3—

On the stone

Memories and epitaphs, kept separately from model reviews.

Epitaphs from visitors

Nobody has written on this stone yet. The first line is yours.

Write an epitaph

One line for Gemini 2.5 Pro Experimental 03-25. No sign-in, no links, published as written.

8–200 characters200

Community discussion

Share the API, setup, and tradeoffs you actually tested. Rate the model once; comment as often as you like.

Used this model? Sign in with Google to rate it and share what worked, what did not, and which API or setup you tested.

No community comments yet.