Gemma 2 27B

Overview

The flagship of the Gemma 2 family from Google. Approaches 70B-class quality on a single 24GB GPU at Q4, with strong multilingual coverage.

When to pick this model

• Self-hosted assistants needing high quality on one consumer GPU
• Multilingual content workflows including French
• Instruction-heavy tasks like classification and structured output
• Workloads that don't require long context
• Replacing 70B models when VRAM is constrained

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	16 GB
Q5_K_M	19 GB
Q8_0	29 GB
FP16 (no quantization)	54 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

Published benchmark scores

Benchmark	Score
MMLU	75.2
HellaSwag	89.5
HumanEval	51.8

Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.

Strengths

• Quality close to 70B-class models
• Runs in roughly 16GB VRAM at Q4
• Strong instruction following
• Robust multilingual output

Limitations

• 8k context is a major handicap in 2025
• Gemma license is less permissive than Apache 2.0
• No vision in this checkpoint

Architecture & training

Architecture: Dense Transformer · Gemma 2 27B · logit-softcapping

Training: 13T tokens. The largest in the Gemma 2 family.

Quick start

ollama run gemma2:27b

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.