Atlas

Models

← All models

Gemini 2.0 Flash

Google DeepMind · 2025-02-05

Gemini 2.0 Flash (`gemini-2.0-flash`) was Google's auto-updated alias for the generally available Gemini 2.0 Flash model, released on February 5, 2025 and pointing to the stable `gemini-2.0-flash-001` version. It accepts text, images, audio, and video with a 1,048,576-token input limit and returns text; experimental and Thinking Experimental variants used separate identifiers.

Benchmark scores

BenchmarkScore
AI2D (OpenVLM)83.1
Artificial Analysis Intelligence Index10.7
Capability130.5
ChessBench Accuracy38.6
ChessBench Coherence27.9
ChessBench Elo348
Confabulations Leaderboard Confabulation Rate24.3
Confabulations Leaderboard Non-Response Rate29.4
Confabulations Leaderboard Weighted Score26.9
Elimination Game TrueSkill Mu2.0
EnigmaEval0.6
Global MMLU Lite83.4
HallusionBench (OpenVLM)58.0
HardGeoBench <=100mi40.0
HardGeoBench <=10mi26.3
HardGeoBench Avg Distance1128
HardGeoBench Country Accuracy51.3
Humanity's Last Exam5.1
Humanity's Last Exam (Text-Only)4.7
LiveCodeBench (Artificial Analysis)21.0
MATH90.9
MATH-500 (Artificial Analysis source)91.1
MATH-Vision41.3
MathVista (mini)73.1
MathVista (OpenVLM)70.4
MEDIC (open-ended Elo)1579
MM-Vet (OpenVLM)73.6
MMMU Validation (OpenVLM)69.9
MMStar (OpenVLM)69.4
MRCR v2 (8-needle) - 128K average19.0
Loading Atlas data…