Gemma 2B benchmark results
5 results on 4 benchmarks from 4 sources.
Open weights Released: Feb 21, 2024 Updated Oct 23, 2026
Also listed as: gemma-2b · Gemma-2B · google/gemma-2b
All benchmark results
| Benchmarks | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|
| Open LLM Leaderboard v2 | 7.3% | 4027 | Average of 6 normalised benchmarks; bfloat16; 2.506B params“google/gemma-2b” | Official leaderboard | Mar 13, 2025 |
| MMLU | 42.3% | 82 | Exact match; 5-shot | Paper / tech report via Epoch AI | Feb 21, 2024 |
| MMLU-Pro | 15.8% | 235 | Overall accuracy; data source: TIGER-Lab; 2B“Gemma-2B” | Official leaderboard | Mar 11, 2026 |
| HumanEval / HumanEval+ | 25.0% | 101 | HumanEval (base tests), pass@1 greedy“gemma-2b” | Official leaderboard | Dec 26, 2024 |
| HumanEval / HumanEval+ | 20.7% | 103 | HumanEval+ (extended tests), pass@1 greedy“gemma-2b” | Official leaderboard | Dec 26, 2024 |
Where Gemma 2B ranks
Primary setting, one best-sourced score per model (official / independent first); neighbours in brackets.
| Benchmark | Score | Rank | Nearby models | #1 |
|---|---|---|---|---|
| Open LLM Leaderboard v2 | 7.3% | #232 / 294 | deepseek-moe-16b-base 7.5% · stablelm-3b-4e1t 7.3% · OLMoE-1B-7B-0924 7.3% · Qwen2-0.5B 7.2% | Qwen2.5-72B48.0% |
| HumanEval / HumanEval+ | 20.7% | #104 / 122 | StarCoderBase-7B 21.3% · codegemma-2b 20.7% · CodeGen2-7B 17.7% · Gemma 1.1 2B IT 17.7% | O1 Preview (Sept 2024)89.0% |
| MMLU | 42.3% | #135 / 158 | InstructGPT 175B 43.9% · Baichuan1-7B 42.3% · Qwen2.5-Coder-0.5B 42.0% · CodeQwen1.5-7B 40.5% | GPT-4o (Nov 2024)88.1% |
| MMLU-Pro | 15.8% | #231 / 243 | SmolLM2-1.7B 18.3% · Qwen2-0.5B-Instruct 15.9% · Gemma-2-2B-it 15.6% · Qwen2-0.5B 15.0% | Gemini 3.1 Pro91.2% |
Other Google models
Gemini 2.5 Pro (Jun 2025)Gemini 3 ProGemini 3.1 ProGemini 1.5 Pro (Sept 2024)Gemini 1.5 Flash (Sep 2024)Gemini 3 FlashGemini 3.8 FlashGemma 2 27BGemma 2 9BGemma 3 27BGemini 1.5 Flash (May 2024)Gemini 1.5 Pro (May 2024)
FAQ
Where does Gemma 2B rank best?
Open LLM Leaderboard v2: #232 of 294 (7.3%); #1 there is Qwen2.5-72B (48.0%).
Where do these Gemma 2B scores come from?
From EvalPlus leaderboard, Epoch AI Benchmarking Hub, MMLU-Pro leaderboard (TIGER-Lab), and Hugging Face Open LLM Leaderboard v2 (archived). Each row links to the original leaderboard, dataset or report, and vendor-reported numbers are labelled.