What is Gemma 3 12B Instruct good for?

Question answering, summarising and image understanding on your own hardware. Open weights under the Gemma Terms of Use, small enough for a single consumer GPU.

When is Gemma 3 12B Instruct not a good fit?

Video or audio input, and structured output. The Gemma Terms of Use are a custom license rather than Apache 2.0, and the Gemma 4 family knows more recent events.

What is the cheapest way to run Gemma 3 12B Instruct?

Hosted, unless you push serious volume. Novita charges $0.05 in / $0.10 out per 1M tokens. The cheapest rental that fits is 1x RTX 4070 at $72 a month, which costs the same as about 1B tokens a month on that API.

Can I self-host Gemma 3 12B Instruct?

Yes. The weights are released under the Gemma Terms of Use. At 4-bit it needs about 10 GB of GPU memory, which starts at roughly $72 a month on the cheapest rental that fits.

Open weights · Gemma Terms of Use

Gemma 3 12B Instruct

The 12B instruction-tuned model in Google's open-weight Gemma 3 family, sized to run on a single consumer GPU.

Jump to: Specs Pricing Self-hosting Similar models FAQ More from Google Cloud

Visit website Compare vs other LLMs →

Cheapest via Novita

Input: $0.05 /1M tokens
Output: $0.10 /1M tokens

2 providers

Key Specifications

Context window: 128K tokens
Max output: 128K tokens
Knowledge cutoff: Aug 2024
Released: Mar 2025
Parameters: 12.2B
Inputs: Text, image

Capabilities Show details: Function calling Connect to external tools, APIs, and systems.

Hosted API pricing

Input tokens M Output tokens M

Provider	Input /1M tokens	Output /1M tokens	Cost at 10M in + 2M out
Novita	$0.05	$0.10	$0.70	View
Google Cloud Creator	Open weights, no hosted price			View

Heads up: Base-tier, on-demand rates per 1M tokens; cached, batch and long-context tiers excluded. A provider may serve a shorter context or a quantized build than the creator's release. Verify before provisioning. More on how we price.

Estimated cost to self-host

Gemma 3 12B Instruct needs about 10 GB of GPU memory at 4-bit with 32K context. The cheapest rental that fits is 1× RTX 4070 at about $72 a month.

That costs the same as roughly 1B tokens a month on Novita's API. Self-hosting is more expensive below that volume.

Fits on 1 Single node ≤ 8

Precision	Memory	Cheapest fit (32K context)(Full 128K context)	Cost /mo	Break-even vs API
4-bitINT4 / FP4	10 GB13 GB	1× RTX 4070	$72$79	1B tokens /mo1B tokens /mo
8-bitFP8 / INT8	16 GB18 GB	2× RTX 3060	$86$144	1B tokens /mo2B tokens /mo
16-bitFP16 / BF16	29 GB35 GB	2× RTX 4060 Ti	$158$173	3B tokens /mo3B tokens /mo

Estimates based on median on-demand rates for Nvidia GPUs. Memory is weights plus KV cache for one request, using FP8 KV cache where supported and FP16/BF16 otherwise. Break-even assumes a 5:1 input-to-output ratio. No guarantee of runtime support, usable performance, or that a matching quantized build exists. How we estimate costs.

Similarly priced models

The models nearest Gemma 3 12B Instruct by blended rate, each at its own cheapest provider.

Model	Blended /1M	Input /1M	Output /1M	Context	Cutoff	vs Gemma 3 12B Instruct
Llama 3.1 8B Meta	$0.025	$0.02	$0.05	131K	Dec 2023	−57%
DeepSeek OCR-2 DeepSeek	$0.03	$0.03	$0.03	8K		−49%
Qwen3.7-Flash Alibaba Cloud	$0.0467	$0.03	$0.13	1M		−20%
GPT-OSS-20B OpenAI	$0.0483	$0.03	$0.14	131K	Jun 2024	−17%
Command R7B Cohere	$0.0583	$0.04	$0.15	128K		0%
Gemma 3 12B Instruct This model Google Cloud	$0.0583	$0.05	$0.10	128K	Aug 2024
Llama 3.2 1B Meta	$0.0583	$0.03	$0.20	128K	Dec 2023	0%
GPT-OSS-120B OpenAI	$0.0633	$0.04	$0.18	131K	Jun 2024	+9%
Nemotron 3 Nano 30B A3B Nvidia	$0.075	$0.05	$0.20	262K	Jun 2025	+29%
Llama 3 8B Meta	$0.0833	$0.05	$0.25	8K	Mar 2023	+43%
Qwen3-Coder-30B-A3B Alibaba Cloud	$0.0917	$0.06	$0.25	262K		+57%

Prices are USD per 1M tokens at each model's cheapest listed provider. Blended is the cost of 10M input plus 2M output tokens, spread over the 12M.

Lyceum

NVIDIA H200 from $2.20/hr, B200 from $3.40/hr, B300 from $3.80/hr

Per-second billing with no minimum commitment or vendor lock-in

Scale to 8,000 GPUs over InfiniBand for distributed AI training

View Pricing

Frequently Asked Questions

What is Gemma 3 12B Instruct good for?: Question answering, summarising and image understanding on your own hardware. Open weights under the Gemma Terms of Use, small enough for a single consumer GPU.
When is Gemma 3 12B Instruct not a good fit?: Video or audio input, and structured output. The Gemma Terms of Use are a custom license rather than Apache 2.0, and the Gemma 4 family knows more recent events.
What is the cheapest way to run Gemma 3 12B Instruct?: Hosted, unless you push serious volume. Novita charges $0.05 in / $0.10 out per 1M tokens. The cheapest rental that fits is 1x RTX 4070 at $72 a month, which costs the same as about 1B tokens a month on that API.
Can I self-host Gemma 3 12B Instruct?: Yes. The weights are released under the Gemma Terms of Use. At 4-bit it needs about 10 GB of GPU memory, which starts at roughly $72 a month on the cheapest rental that fits.

More from Google Cloud

Model	Context	Cutoff	Input /1M	Output /1M
Gemma 4 26B A4B Open weights	256K	Jan 2025	$0.10	$0.30
Gemma 3 27B Instruct Open weights	128K	Aug 2024	$0.10	$0.30
Gemini 2.5 Flash-Lite	1M	Jan 2025	$0.10	$0.40
Gemma 4 31B Open weights	256K	Jan 2025	$0.13	$0.37
Gemini 3.1 Flash-Lite	1M	Jan 2025	$0.25	$1.50
Gemini 3.5 Flash-Lite	1M	Mar 2026	$0.30	$2.50
Gemini 2.5 Flash	1M	Jan 2025	$0.30	$2.50
Gemini 3 Flash Preview	1M	Jan 2025	$0.50	$3.00