Gemma 2 9B

Gemma 2 9B

Quick answer: Gemma 2 9B is Google's July 2024 mid-size open-weight model, scoring 88.0% on ARC-E, 68.6% on GSM8K, and 81.9% on HellaSwag. It was designed for efficient deployment at 9B scale, but has since been superseded by Gemma 3 12B with a much larger 128K context window.

At a Glance

Where Gemma 2 9B leads

• 88.0% ARC-E — competitive scientific reasoning at 9B scale
• Lower inference cost than 27B variant
• Open weights on Hugging Face
• Efficient architecture

Where it lags

• 8K context window — very restrictive for production use
• Superseded by Gemma 3 12B: 128K context, multimodal, better benchmarks
• Gemma Terms of Use (not Apache 2.0)
• Below Llama 3.1 8B (Apache 2.0, 128K) on most metrics

Best for: Existing Gemma 2 9B integrations; legacy support.

What Gemma 2 9B Is

Gemma 2 9B (released July 31, 2024) was Google's mid-tier Gemma 2 model, positioned between the 2B and 27B variants. Like Gemma 2 27B, its primary limitation is the 8K context window.

For new deployments at the 9-12B scale, Gemma 3 12B is the recommended replacement: 128K context, multimodal vision capability, better benchmarks, and same Gemma Terms of Use license. Llama 3.1 8B (Apache 2.0, 128K context) is also preferred if a fully permissive license is needed.

Specifications

Pricing

Open weights under Gemma Terms of Use — free to self-host.

Public Benchmark Scores

Gemma 2 9B vs Alternatives

For new deployments: Gemma 3 12B (128K, vision, same license) or Llama 3.1 8B (Apache 2.0, 128K, higher GSM8K).

Related Models & Benchmarks

• Compare: Gemma 2 27B · Gemma 3 12B · Llama 3.1 8B Instruct
• Benchmarks: HellaSwag · GSM8K

Frequently Asked Questions

Specs from Google's official Gemma 2 release (July 2024) and Benchgen evaluations. Last updated 2026-07-24.