GoogleUnited States· released Nov 2025

Gemini 3 Pro

Name: Gemini 3 Pro
Author: Google

Provisional: 19 comparable tests so far. Ranked once it reaches 6 across 3 counting categories.

Google, United States. One of 14 models from this lab on the Index.

Third-party publishedLow confidence (thin coverage)

Snapshot September 8, 2026

OpenCharts Index

72.3

Provisional: not enough comparable tests or counting categories to rank yet · 19 comparable tests · 2 counting categories

Drop any single test and the Index lands between 69.4 and 76.9.

6 of 8 hardest tests · share of the frontier

Distance to today's frontier on the hardest tests. Not a measure of intelligence.

What it's known for

Gemini 3 Pro, in 19 results.

Strongest in agentic (78.0) and reasoning (66.6). It trails the frontier most in reasoning, 33.4% behind on average. No comparable results yet in knowledge and long context. Between 25% and 30% behind the best published results, on average.

Strongest category

Agentic

Long-horizon tasks with tools: terminals, desktops, browsers, whole jobs.

78.0

DeepResearch Bench46.3% · 84% of best
Terminal-Bench69.4% · 82% of best
GDPval40.3% · 81% of best
METR Time Horizon3.7 h · 78% of best
Vending-Bench 2$5,478 · 77% of best
APEX-Agents31.5% · 67% of best

Mean of 6 of the category's 8 comparable tests. Counts toward the Index.

Best single result

GPQA Diamond

Answer a PhD-level multiple-choice question in physics, chemistry or biology that a web search will not settle.

92.6%

95.5% of the best (95.8%), above chance · #15 of 78.

Expert-level science questions are the closest thing to asking a specialist colleague. A model that gets them right can check a technical claim instead of echoing it.

Where it trails

ProofBench

Write a complete, rigorous proof, graded for rigor rather than for a final answer.

20.0%

80.0% behind the best published result (100.0%) · #36 of 61.

A proof shows the reasoning itself. A right answer with a wrong argument scores nothing.

Not measured yet

No comparable result yet in these 2 categories. A blank is a blank, never a zero, and it does not lower the Index.

Knowledge
Long context

Covered, but not averaged into the Index:

Coding3 of 7 tests · needs 4 · mean 61.7
Math2 of 5 tests · needs 3 · mean 55.7
Multimodalsingle-test category, shown beside the Index · 93.0
Human preferencesingle-test category, shown beside the Index · 93.8

Every score, with its source

See it at work

What Gemini 3 Pro was asked to do, and how it did.

Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.

Human preference · mean 93.8 across 1 comparable test · a single-test category, shown beside the Index and never averaged into it

Text ArenaLeaderboard

Answer the same prompt as a rival model; real users vote blind on which answer they prefer.

1,486

47% expected win rate against the board leader (1,507)
#13 of 92

board 2026-09-02

Source: Arena

Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.

Head to head

Gemini 3 Pro against whoever you pick.

Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.

Gemini 3 Pro

provisional·Index 72.3B-

0 : 9

wins · 0 ties · 9 shared

GPT-6 Astra

#1·Index 99.0A+

GPT-6 Astra comes out ahead on 9 of the 9 tests both have taken.

Category means

Reasoning
66.6
99.6
Knowledge
—
100.0
Coding
61.7
97.7
Math
55.7
99.8
Agentic
78.0
98.5
Multimodal
93.0
—
Human preference
93.8
—

Shared tests · biggest gaps first

ProofBenchMath
20.0%
99.0%
−79.0
Code Arena (WebDev)Coding
1,439
1,797
−77.5
CritPtReasoning
6.9%
31.7%
−76.8
ARC-AGI-2Reasoning
31.1%
95.0%
−67.3
APEX-AgentsAgentic
31.5%
46.7%
−32.0
WeirdMLCoding
69.9%
92.9%
−24.7
ARC-AGI-1Reasoning
75.0%
98.5%
−23.9
Mock AIME 2024–2025Math
91.4%
100.0%
−8.6
GPQA DiamondReasoning
92.6%
95.8%
−4.5

Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.

Open GPT-6 Astra

Scores

Every test, every source

The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted.

Category means

Reasoning66.6 · 6/6
Knowledge—
Coding61.7 · 3/7
Math55.7 · 2/5
Agentic78.0 · 6/8
Multimodalbeside93.0 · 1/1
Human preferencebeside93.8 · 1/1
Long context—

Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.

GPQA DiamondEpoch-runReasoning

92.6%

95.5% of the best (95.8%), above chance
82th percentile of 78 · n = 198

preview

Gemini 3 Pro

Gemini 3 Pro, in 19 results.

What Gemini 3 Pro was asked to do, and how it did.

Gemini 3 Pro against whoever you pick.

Every test, every source

Compare with its neighbours.

The best models, ranked here, working inside Theo.