Gemma 3 12B IT

google/gemma-3-12b-it

Maker: Google, United States
Released: 2025-03-12
Knowledge cutoff: 2024-08
Family: gemma
Weights: open (gemma)
Tasks: text.chat, vision.understand, vision.extract
Cheapest offering: DeepInfra, about $0.000095 for a typical text.chat call

Try it in the sandbox Use it from the API

Offerings

Provider	Model id at the provider	Status	Price per 1M tokens	Per-unit prices	Limits	Capabilities (level)
Amazon Bedrock	`google.gemma-3-12b-it`	deprecateddeprecated 2027-03-30	Input: $0.09 Output: $0.29 Since 2026-11-03 · models.dev, MIT	none	Context: 131,072 tokens Max output: 8,192 tokens	`input.text` claimed `input.image` claimed `input.pdf` claimed `output.text` claimed `structured_output` claimed `tools` claimed
DeepInfra	`google/gemma-3-12b-it`	ga	Input: $0.05 Output: $0.15 Unverified: sources disagree by more than 5%Since 2026-11-03 · models.dev, MIT	none	Context: 131,072 tokens Max output: 131,072 tokens	`input.text` claimed `input.image` claimed `input.pdf` claimed `output.text` claimed `tools` claimed `structured_output` claimed `tool_choice` claimed
OpenRouter	`google/gemma-3-12b-it`	ga	Input: $0.05 Output: $0.15 Since 2026-11-03 · models.dev, MIT	none	Context: 131,072 tokens Max output: 16,384 tokens	`input.text` claimed `input.image` claimed `input.pdf` claimed `output.text` claimed `tools` claimed `structured_output` claimed

Run it locally

Free on your own device. Memory is the weights, the KV cache at the context shown and about 10% overhead; a longer context needs more. GB are 10⁹ bytes. See local models.

Runtime	Model id in the runtime	Quantization	Needs	Context
llama.cpp	`lmstudio-community/gemma-3-12b-it-GGUF:Q4_K_M`	Q4_K_M	Memory: 11.58 GB at 8,192 tokens GPU memory: 11.58 GB Disk: 7.31 GB Parameters: 11.8B License: gemma	Context: 131,072 tokens
llama.cpp	`lmstudio-community/gemma-3-12b-it-GGUF:Q6_K`	Q6_K	Memory: 14.18 GB at 8,192 tokens GPU memory: 14.18 GB Disk: 9.67 GB Parameters: 11.8B License: gemma	Context: 131,072 tokens
llama.cpp	`lmstudio-community/gemma-3-12b-it-GGUF:Q8_0`	Q8_0	Memory: 17.31 GB at 8,192 tokens GPU memory: 17.31 GB Disk: 12.51 GB Parameters: 11.8B License: gemma	Context: 131,072 tokens
LM Studio	`lmstudio-community/gemma-3-12b-it-GGUF/gemma-3-12b-it-Q4_K_M.gguf`	Q4_K_M	Memory: 11.58 GB at 8,192 tokens GPU memory: 11.58 GB Disk: 7.31 GB Parameters: 11.8B License: gemma	Context: 131,072 tokens
LM Studio	`lmstudio-community/gemma-3-12b-it-GGUF/gemma-3-12b-it-Q6_K.gguf`	Q6_K	Memory: 14.18 GB at 8,192 tokens GPU memory: 14.18 GB Disk: 9.67 GB Parameters: 11.8B License: gemma	Context: 131,072 tokens
LM Studio	`lmstudio-community/gemma-3-12b-it-GGUF/gemma-3-12b-it-Q8_0.gguf`	Q8_0	Memory: 17.31 GB at 8,192 tokens GPU memory: 17.31 GB Disk: 12.51 GB Parameters: 11.8B License: gemma	Context: 131,072 tokens
MLX	`mlx-community/gemma-3-12b-it-4bit`	4bit	Memory: 12.38 GB at 8,192 tokens GPU memory: 12.38 GB Disk: 8.03 GB Parameters: 11.8B License: gemma	Context: 131,072 tokens
Ollama	`gemma3:12b`	Q4_K_M	Memory: 12.51 GB at 8,192 tokens GPU memory: 12.51 GB Disk: 8.15 GB Parameters: 11.8B License: gemma	Context: 131,072 tokens

Quality scores

Basis	Score (0–1)	Raw	Measured	Source
`epoch:gpqa_diamond`	0.409	0.3946	2026-08-28	Epoch AI, CC-BY-4.0, normalized by Nanner Tech

Aliases

Other names the API accepts for this model.

google.gemma-3-12b-it

Use it from the API

Look this model up, or ask for the best model for its task on the providers that offer it. See the quickstart and the API reference.

Look up this model

curl 'https://rhumbdata.com/v1/models/google/gemma-3-12b-it' \
  -H "Authorization: Bearer $RHUMB_KEY"

Resolve text.chat on its providers

curl 'https://rhumbdata.com/v1/resolve?task=text.chat&providers=bedrock,deepinfra,openrouter' \
  -H "Authorization: Bearer $RHUMB_KEY"

Built from catalog version 2026-11-04.3 of 2026-11-04. Prices are list prices per 1M tokens unless marked otherwise; every score shows its source's notice (see Attributions).