Atlas

Models

← All models

Gemini 2.0 Flash Thinking Experimental 01-21

Google DeepMind · 2025-01-21

Gemini 2.0 Flash Thinking Experimental 01-21 is Google's January 21, 2025 preview snapshot of the test-time-compute model behind Gemini 2.0 Flash Thinking. It accepts text, image, audio, and video across a 1,048,576-token context, produces text, and performs thinking intrinsically rather than through a user-selectable thinking-budget control.

Benchmark scores

BenchmarkScore
Agentic Tool Use (Chat)57.4
Agentic Tool Use (Enterprise)63.2
Aider Polyglot18.2
Arabic1120
Artificial Analysis Coding Index24.1
Artificial Analysis Intelligence Index13.3
Capability133.6
Chinese1060
Coding1108
Confabulations Leaderboard Confabulation Rate14.9
Confabulations Leaderboard Non-Response Rate10.0
Confabulations Leaderboard Weighted Score12.4
Elimination Game TrueSkill Mu2.8
EnigmaEval1.1
Epoch Capabilities Index (ECI)136
Fiction.liveBench52.8
Humanity's Last Exam6.6
Humanity's Last Exam (Preview)7.2
Humanity's Last Exam (Text Only)6.5
Humanity's Last Exam (Text-Only)7.1
Humanity's Last Exam Text Only (Preview)7.0
LiveBench66.9
LiveCodeBench (Artificial Analysis)32.1
LLM Creative Story-Writing Benchmark7.4
LLM Divergent Thinking Repeat Rate0.2
LLM Divergent Thinking Score4.4
MASK49.5
MATH-50084.6
MATH-500 (Artificial Analysis source)94.4
MMLU Pro79.8
Loading Atlas data…