Gemini 3 Flash Preview High

Released2025-12-17
Price per 1M$0.50 in · $3.00 out
Overall score55.4 · rank 194
Context window1,048,576 tokens
ProviderGoogle
WeightsProprietary
ConfigurationHigh

Scores

DimensionScoreRankBenchmarks
Overall: Balanced 55.4 194 22
Overall: Coding 53.9 206 16
Overall: Reasoning 56.8 169 10
Overall: Agentic work 55.1 199 15
Overall: Frontend 50.0 216 5
Coding 55.4 172 3
Agents 54.3 156 7
Reasoning 57.8 164 5
Math 55.3 116 2
Knowledge 57.0 59 3
Frontend 45.5 146 1
Multimodal 60.4 28 1
Value vs price+5.0131

Evidence

Coding 55.4 ±3.3

BenchmarkResultSourceDate
SWE-bench Verified (Vals run) 75.0% Vals AI 2026-09-01
LiveCodeBench (Vals run) 85.6% Vals AI 2026-09-01
ProgramBench 0.0% Vals AI 2026-11-09

Agents 54.3 ±1.8

BenchmarkResultSourceDate
Terminal-Bench 2.1 (Vals run) 53.9% Vals AI 2026-10-01
tau2-bench Banking Knowledge 27.3% tau2-bench 2026-03-02
Tax Agent Bench 11.8% Vals AI 2026-11-09
tau2-bench Airline 82.5% tau2-bench 2026-03-02
DeepResearch Bench 47.9% Epoch AI 2026-11-11
tau2-bench Telecom 91.2% tau2-bench 2026-03-02
tau2-bench Retail 76.8% tau2-bench 2026-03-02

Reasoning 57.8 ±2.2

BenchmarkResultSourceDate
ARC-AGI-2 (ARC Prize verified) 33.6% ARC Prize 2026-11-11
ARC-AGI-1 (ARC Prize verified) 84.7% ARC Prize 2026-11-11
GPQA Diamond 89.4% Epoch AI 2026-08-06
GPQA Diamond (Vals run) 87.9% Vals AI 2026-09-01
Chess Puzzles 40.0% Epoch AI 2026-10-13
Mystery Game Puzzles 20.0% Epoch AI 2026-08-05

Math 55.3 ±2.3

BenchmarkResultSourceDate
ProofBench 15.0% Epoch AI 2026-11-11
OTIS Mock AIME 2024-2025 95.6% Epoch AI 2026-08-06

Knowledge 57.0 ±6.3

BenchmarkResultSourceDate
SimpleQA Verified 66.8% Epoch AI 2026-08-27
MMLU-Pro (Vals run) 88.6% Vals AI 2026-09-01
Legal Research Bench 18.3% Vals AI 2026-11-09

Frontend 45.5 ±0.8

BenchmarkResultSourceDate
Vibe Code Bench v1.1 20.2% Vals AI 2026-11-09

Multimodal 60.4 ±2.7

BenchmarkResultSourceDate
MMMU-Pro (Vals run) 87.6% Vals AI 2026-09-01

Names matched to this model

gemini-3-flash-preview-high

Compare Gemini 3 Flash Preview High with other models