Atlas

Models

← All models

Gemini 2.0 Pro Experimental 02-05

Google DeepMind · 2025-02-05

Gemini 2.0 Pro Experimental 02-05 is Google's versioned February 5, 2025 public-preview snapshot of Gemini 2.0 Pro, positioned for coding and complex prompts. The proprietary hosted model accepts text, images, audio, and video across a 2,097,152-token input context, returns text, supports tool use, and has no distinct thinking mode.

Benchmark scores

BenchmarkScore
Agentic Tool Use (Chat)57.9
Agentic Tool Use (Enterprise)61.5
Aider Polyglot35.6
Artificial Analysis Coding Index25.5
Artificial Analysis Intelligence Index11.8
Capability133.1
Confabulations Leaderboard Confabulation Rate15.8
Confabulations Leaderboard Non-Response Rate21.0
Confabulations Leaderboard Weighted Score18.4
CorpFin v243.4
CorpFin v2 - Exact Pages51.2
CorpFin v2 - Max Fitting Context38.7
CorpFin v2 - Shared Max Context40.4
Elimination Game TrueSkill Mu3.1
EnigmaEval0.7
Fiction.liveBench41.7
Humanity's Last Exam (Preview)7.1
Humanity's Last Exam (Text-Only)6.8
Humanity's Last Exam Text Only (Preview)6.7
LiveBench65.1
LiveCodeBench (Artificial Analysis)34.7
MASK48.7
MATH Level 583.5
MATH-500 (Artificial Analysis source)92.3
MATH-Vision48.1
MMLU Pro80.5
PGG-Bench TrueSkill Mu11.1
SciCode (Artificial Analysis)31.3
Step Game TrueSkill Mu1.4
TaxEval v268.2
Loading Atlas data…