Gemma 3 12B
Open Gemma instruction model for efficient chat and self-hosted deployments
Model facts
- Context window
- 131K tokens
- Maximum output
- 16K tokens
- Cheapest paid input
- $0.05 per 1M tokens
- Cheapest paid output
- $0.1 per 1M tokens
- Open weights
- yes
- Providers
- 8
Capabilities and modalities
tool calling, structured output, attachments, temperature control, open weights, text, image, pdf.
Local hardware estimate
At 4K context: Q4_K_M 8.4 GiB, Q8_0 14.1 GiB, F16 24.8 GiB working memory.
- Parameters
- 12.2 billion
- Architecture
- gemma3
- Model license
- gemma
Planning estimate adapted from the MIT-licensed whichllm estimator using metadata from Hugging Face; not a vendor minimum requirement.
Artificial Analysis benchmarks
Compare this model on the full LLM benchmark leaderboard.
| Configuration | AAI | Coding | Math | Output tokens/s |
|---|---|---|---|---|
| default | 5.5 | 5.8 | 18.3 | not measured |
Observed price history
- 2026-08-24: $0.05 input / $0.1 output per 1M tokens
API providers
- Nvidia — model id google/gemma-3-12b-it; free input / free output per 1M tokens; provider documentation
- Amazon Bedrock (Google) — model id google.gemma-3-12b-it; $0.05 input / $0.1 output per 1M tokens; provider documentation
- Helicone (Google) — model id gemma-3-12b-it; $0.05 input / $0.1 output per 1M tokens; provider documentation
- NovitaAI — model id google/gemma-3-12b-it; $0.05 input / $0.1 output per 1M tokens; provider documentation
- OpenRouter — model id google/gemma-3-12b-it; $0.05 input / $0.15 output per 1M tokens; provider documentation
- Kilo Gateway (Google) — model id google/gemma-3-12b-it; $0.05 input / $0.15 output per 1M tokens; provider documentation
- NanoGPT — model id unsloth/gemma-3-12b-it; $0.272 input / $0.272 output per 1M tokens; provider documentation
- Neon — model id gemma-3-12b; $0.15 input / $0.5 output per 1M tokens; provider documentation