Google
Efficient

Gemma 2 9B

Open

Mid-size Gemma 2, popular open model in the 9B class for on-prem chat and fine-tuning.

Gemma 2 9B is a efficient AI model from Google. It costs $0.0900 per million input tokens and $0.0900 per million output tokens (blended $0.0900/M), with a 8,000-token context window.

RatesUSD per million tokens. Blend is 70% input, 30% output, our calculation.

INPUT
$0.0900/M
per million input tokens
OUTPUT
$0.0900/M
per million output tokens
BLENDED 70/30
$0.0900/M
unchanged since 3 May
CONTEXT
8,000
tokens

Hand-maintained from Google's published pricing; no check date on record.

What it is good at
  • Single-GPU friendly
  • Open weights
  • Wide ecosystem
Typical use cases
  • On-prem chat
  • Fine-tune base

Benchmarks

vs. best public score
MMLU71.3%
Multitask academic knowledge across 57 subjects. Source: Google DeepMind (self-reported), MMLU (5-shot), read 2026-11-10.
Graduate-level science questions, "Google-proof". No source recorded yet.
MATH45%
High-school competition math problems. No source recorded yet.
HumanEval40.2%
Python function synthesis from docstrings. Source: Google DeepMind (self-reported), HumanEval pass@1, read 2026-11-10.
LMArena Elo1267 Elo
Crowd-sourced head-to-head preference Elo rating. Source: LMArena (independent), LMArena Text Arena score, read 2026-11-10.
Hand-curated from each provider's published reports and public leaderboards. Methodology varies across sources, treat as directional rather than authoritative.

More from Google

See all 23 →

Frequently asked questions

How much does Gemma 2 9B cost?

Gemma 2 9B costs $0.0900 per million input tokens and $0.0900 per million output tokens, for a blended reference rate of $0.0900 per million tokens.

What is Gemma 2 9B's context window?

Gemma 2 9B supports up to 8,000 tokens of context in a single request.

What is Gemma 2 9B best for?

Gemma 2 9B is well suited to Single-GPU friendly, Open weights and Wide ecosystem.

Who makes Gemma 2 9B?

Gemma 2 9B is developed and served by Google. It was released in Jun 2024.

Terms used on this page