---
title: "Gemma 3 27B Benchmarks - RankedAGI"
description: "Detailed benchmark and metadata record for Gemma 3."
source: "https://rankedagi.com/models/gemma-3-27b"
---

# Gemma 3

| Field | Value |
| --- | --- |
| Organization | Google |
| License | Open Source |
| Size | 27B |
| Version | Latest |
| Released | 2025-03-12 |
| Model ID | gemma-3-27b-it |
| Last updated | 2026-04-27T17:12:46.239Z |

## Benchmarks

| Benchmark | Category | Value | Description | Source |
| --- | --- | --- | --- | --- |
| RankedAGI Coding | coding | 24.5% | RankedAGI Coding Score |  |
| RankedAGI Agentic | Agents | 23.2% | RankedAGI Agentic Score |  |
| LiveCodeBench v6 | coding | 14.0% | LiveCodeBench v6 - Evaluation of Large Language Models for Code |  |
| ChatArena (LMSYS) | coding | 1244 | ChatArena (LMSYS) Coding ELO Score with Style Control |  |
| Aider Polyglot | coding | 4.9% | Aider Polyglot Code Completion Benchmark |  |
| RankedAGI Reasoning | reasoning | 30.0% | RankedAGI Reasoning Score |  |
| Humanity's Last Exam | reasoning | 5.0% | Multidisciplinary Reasoning (no tools) |  |
| GPQA Diamond | reasoning | 42.4% | Generalized Prefix Question Answering Score (Reasoning) PhD Level Reasoning |  |
| Text Arena | general | 1339 | ChatArena (LMSYS) ELO Score | https://arena.ai/leaderboard/text |
| AIME 2025 | math | 21.0% | AIME 2025 Competition Math |  |
| HumanEval | coding | 87.8% | HumanEval coding benchmark |  |
| NYT Connections Extended | reasoning | 11.8% | NYT Connections Extended Version | https://github.com/lechmazur/nyt-connections/ |
| MMLU Pro | knowledge | 67.5% | Graduate-level visual problem-solving |  |
| MMLU | knowledge | 76.9% | Image Reasoning |  |
| MMMU | imaging | 64.9% | Multimodal Understanding College-level visual problem-solving |  |
| Hughes Hallucination Evaluation | safety | 3.0% | Computed using Vectara's Hallucination Evaluation Model, also known as HHEM. This evaluates how often an LLM introduces hallucinations when summarizing a document. | https://github.com/vectara/hallucination-leaderboard |
| Instruction Following Evaluation | instruction-following | 32.0% | Instruction Following Evaluation |  |
| LiveBench Coding 25.4 | coding | 25.4% | LiveBench Coding Score 2025-04 | https://livebench.ai/ |
| MMMU-Pro | imaging | 48.0% | Multimodal Understanding |  |
| RankedAGI Math | math | 41.2% | RankedAGI Math Score |  |
| RankedAGI Overall | general | 34.4% | Overall RankedAGI score |  |
| 𝜏²-Bench Telecom | Agents | 11.0% | Agentic Tool Use |  |
| SvelteBench old | coding | 11.1% | SvelteBench - Benchmark for Svelte (Old v1) | https://khromov.github.io/svelte-bench/v1/v1-benchmark-results-merged.html |

## Model Links

| Title | URL |
| --- | --- |
| Release | https://blog.google/technology/developers/gemma-3/ |