Google DeepMindopen-weight
Released: 2024-07-31Gemma 2 2B
Ultra-compact mobile and edge AI model running smoothly on CPU, laptop GPUs, and embedded hardware.
Intelligence Index
58.0/ 100
Calibrated multi-domain compositeThroughput Speed
180tok/sec
Output streaming throughputTime To First Token
160ms
Initial chunk server latencyLiveBench Contam-Free
56.4%
September 2026 suite scoreLiveBench Multi-Domain Evaluation Breakdown
Monthly refreshed contamination-free test setsLogical Reasoning58.0%
Coding & Repo Repair52.0%
Mathematics (AIME)54.5%
Data Analysis & Tables55.0%
Language & Comprehension58.0%
Instruction Following60.0%
Sourced Benchmark Evaluations (51)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| LiveBench (Continuous) | Active | 56.4% | 94 | 2025-02-01 | independent | Source ↗ |
| ARC-AGI-3 | Active | 36.4% | 93 | 2024-09-04 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 43.7% | 75 | 2024-08-28 | independent | Source ↗ |
| FrontierMath | Active | 15.4% | 98 | 2024-08-25 | independent | Source ↗ |
| PutnamBench | Active | 28.6% | 90 | 2024-08-24 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 35.4% | 20 | 2024-08-22 | independent | Source ↗ |
| Omni-MATH | Deprecated | 40.7% | 10 | 2024-08-22 | independent | Source ↗ |
| Soohak | Active | 41.8% | 98 | 2024-08-20 | independent | Source ↗ |
| FrontierCode | Active | 30.2% | 98 | 2024-08-20 | independent | Source ↗ |
| AIME | Nearing Saturation | 52.3% | 75 | 2024-08-18 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 51% | 75 | 2024-08-18 | independent | Source ↗ |
| Humanity's Last Exam | Active | 24.6% | 93 | 2024-08-18 | independent | Source ↗ |
| OlympiadBench | Active | 45.1% | 80 | 2024-08-17 | independent | Source ↗ |
| MathArena | Active | 44% | 99 | 2024-08-16 | independent | Source ↗ |
| Terminal-Bench | Active | 36.4% | 80 | 2024-08-16 | independent | Source ↗ |
| FireBench | Active | 50.4% | 95 | 2024-08-16 | independent | Source ↗ |
| SimpleQA | Active | 30.2% | 90 | 2024-08-15 | independent | Source ↗ |
| CodeContests | Deprecated | 33.8% | 10 | 2024-08-15 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 49.2% | 80 | 2024-08-14 | independent | Source ↗ |
| SWE-bench Pro | Active | 32.2% | 90 | 2024-08-14 | independent | Source ↗ |
| MultiPL-E | Deprecated | 47.8% | 10 | 2024-08-13 | independent | Source ↗ |
| LiveCodeBench | Saturated | 42.6% | 39 | 2024-08-12 | independent | Source ↗ |
| MATH | Saturated | 53.9% | 25 | 2024-08-12 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 40.8% | 80 | 2024-08-12 | independent | Source ↗ |
| Mercury | Deprecated | 45.8% | 10 | 2024-08-11 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 53.4% | 75 | 2024-08-11 | independent | Source ↗ |
| CyberSecEval | Deprecated | 46.8% | 10 | 2024-08-10 | independent | Source ↗ |
| TabMWP | Saturated | 50.6% | 25 | 2024-08-10 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 40.6% | 25 | 2024-08-09 | independent | Source ↗ |
| MMLU-Redux | Saturated | 56.8% | 25 | 2024-08-09 | independent | Source ↗ |
| SWE-bench | Deprecated | 35.4% | 10 | 2024-08-08 | independent | Source ↗ |
| MGSM | Nearing Saturation | 52.3% | 65 | 2024-08-08 | independent | Source ↗ |
| TruthfulQA | Saturated | 51% | 25 | 2024-08-07 | independent | Source ↗ |
| AGIEval | Saturated | 48.2% | 25 | 2024-08-07 | independent | Source ↗ |
| SuperGLUE | Saturated | 57.1% | 25 | 2024-08-06 | independent | Source ↗ |
| GSM-Hard | Deprecated | 52.8% | 10 | 2024-08-06 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 6.2% | 75 | 2024-08-05 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 54.9% | 25 | 2024-08-05 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 48.9% | 23 | 2024-08-05 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 53% | 65 | 2024-08-04 | independent | Source ↗ |
| IFEval | Deprecated | 58.8% | 25 | 2024-08-04 | independent | Source ↗ |
| WinoGrande | Saturated | 58.2% | 28 | 2024-08-04 | independent | Source ↗ |
| GPQA | Nearing Saturation | 51% | 78 | 2024-08-04 | independent | Source ↗ |
| CommonsenseQA | Saturated | 58.8% | 28 | 2024-08-03 | independent | Source ↗ |
| HumanEval+ | Saturated | 56.2% | 25 | 2024-08-03 | independent | Source ↗ |
| GLUE | Saturated | 60.5% | 28 | 2024-08-02 | independent | Source ↗ |
| MBPP | Saturated | 55.1% | 25 | 2024-08-02 | independent | Source ↗ |
| GSM8K | Saturated | 59.4% | 25 | 2024-08-02 | independent | Source ↗ |
| MMLU | Saturated | 59.2% | 25 | 2024-08-01 | independent | Source ↗ |
| HumanEval | Saturated | 59.3% | 25 | 2024-08-01 | independent | Source ↗ |
| ARC-AGI | Saturated | 52.6% | 25 | 2024-08-01 | independent | Source ↗ |
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run gemma-2-2b
# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "gemma-2-2b",
"messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
}'Evaluation & Sourcing Notes
Scores listed for Gemma 2 2B represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.