Atlas

Models

← All models

Gemini 2.0 Flash Experimental

Google DeepMind · 2024-12-11

Gemini 2.0 Flash Experimental was Google's public-preview `gemini-2.0-flash-exp` endpoint, released December 11, 2024 with text, image, audio, and video input, text output for all developers, native tool use, and a 1,048,576-token context. It was separate from the stable `gemini-2.0-flash-001` and the Flash Thinking Experimental models; because Google's experimental endpoints are not stable, the same model ID was later used for public native-image-generation experimentation in March 2025.

Benchmark scores

BenchmarkScore
Agentic Tool Use (Chat)53.3
Agentic Tool Use (Enterprise)60.1
AidanBench1872
Aider Polyglot22.2
Arabic1090
BigCodeBench Complete59.9
BigCodeBench Instruct45.9
BigCodeBench-Hard Complete33.8
BigCodeBench-Hard Instruct23.0
Capability131.8
Chinese1115
Coding1111
Epoch Capabilities Index (ECI)135
Humanity's Last Exam (Preview)5.2
Humanity's Last Exam Text Only (Preview)4.9
Instruction Following86.6
Japanese1064
LLM Divergent Thinking Repeat Rate2.6
LLM Divergent Thinking Score4.7
MATH-50089.0
MortgageTax - Numerical Extraction (Vals all 1,258 documents)64.2
MortgageTax - Semantic Extraction (Vals all 1,258 documents)56.5
MortgageTax (Vals all 1,258 documents)60.4
Public Goods Game Average Contribution45.2
SimpleBench18.9
Spanish1119
Step Game TrueSkill Mu1.0
TaxEval v267.7
TaxEval v2 - Correctness56.1
TaxEval v2 - Stepwise Reasoning79.4
Loading Atlas data…