Gemma 3 12B
Language modelActive
Gemma 3 12B
- Released
- March 10, 2025
- Data date
- September 8, 2026
- Price
- Input · Output / 1M tokens
- Input
- Text, Image
- Output
- Text
Gemma 3 12B is distinguished by carrying the image understanding of Gemma 3 27B into a mid-sized dense checkpoint with 12 billion parameters and a 128,000-token context. Google reports 12 trillion training tokens and more than 140 training languages.
The 12B instruction-tuned weights process text and images and generate text. The weight page documents the exact google/gemma-3-12b-it path for local use with Transformers and vLLM.
Specifications and access
| Specification | Value and source |
|---|---|
| Weights | Open weightsSource |
| Access | API, LocalSource |
| Parameters | 12BSource |
| Context window | 128,000 tokensSource |
| Knowledge cutoff | August 2024Source |
| Release | 2025-03-12Source |
| Parameters | 12 BSource |
| Input | Text, ImageSource |
| Output | TextSource |
| Architecture | Decoder-only transformer with multimodal vision encoder; instruction-tuned variantSource |
| License | Gemma Terms of UseSource |
| Training tokens | 12,000,000,000,000 tokensSource |
| Knowledge cutoff | 2024-08Source |
| Languages | 140+Source |
Page 1 of 6
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
GPQA Diamond (Show measurement, test conditions, and source)
- Source value
- 40.9
- Score
- 40.9
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
SimpleQA (Show measurement, test conditions, and source)
- Source value
- 6.3
- Score
- 6.3
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
FACTS Grounding (Show measurement, test conditions, and source)
- Source value
- 75.8
- Score
- 75.8
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; -; instruction-tuned 12B
BIG-Bench Hard (Show measurement, test conditions, and source)
- Source value
- 85.7
- Score
- 85.7
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
BIG-Bench Extra Hard (Show measurement, test conditions, and source)
- Source value
- 16.3
- Score
- 16.3
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
IFEval (Show measurement, test conditions, and source)
- Source value
- 88.9
- Score
- 88.9
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
MMLU Pro (Show measurement, test conditions, and source)
- Source value
- 60.6
- Score
- 60.6
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
LiveCodeBench (Show measurement, test conditions, and source)
- Source value
- 24.6
- Score
- 24.6
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
Bird-SQL dev (Show measurement, test conditions, and source)
- Source value
- 47.9
- Score
- 47.9
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; -; instruction-tuned 12B
Math (Show measurement, test conditions, and source)
- Source value
- 83.8
- Score
- 83.8
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
HiddenMath (Show measurement, test conditions, and source)
- Source value
- 54.5
- Score
- 54.5
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 0-shot; instruction-tuned 12B
MBPP (Show measurement, test conditions, and source)
- Source value
- 73
- Score
- 73
- Metric
- percent
- Unit
- %
- Category
- vendor-evaluation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Gemma 3 model-card table; 3-shot; instruction-tuned 12B
Page 1 of 6
Our tests
SVG short test
One SVG prompt and fixed output settings. Captures are shown only with 0 reported reasoning tokens or documented lack of reasoning controls. Identical test conditions do not make the models' internal compute identical.
Earlier tests
Pelican on a bicycle
A community classic for free-form SVG drawing.
Gemma 3 12B
Output withheld for safety
Reasoning (requested): not specified
Current attempt
- Run date
- Oct 2, 2026
- API provider
- openrouter
- Requested API model
- google/gemma-3-12b-it
- API model reported in response
- google/gemma-3-12b-it
- Reasoning setting
- not specified
- Pinned endpoint
- DeepInfra | google/gemma-3-12b-it
- Temperature
- Provider default
- Token limit including reasoning
- 8,192 tokens
- Test protocol
- community-visual-tests-v1
- Technical checks
- Failed
- Response time
- 45.7 s
Not yet visually reviewed. A successful recording does not confirm correct physics or full compliance with the prompt.
Earlier attempts (1)
- Oct 1, 2026Output withheld for safetyReasoning setting: not specified
Prompt and test conditions
Generate an SVG of a pelican riding a bicycle
Token limit including reasoning: 8,192 tokens
Task origin (Simon Willison)Popularity
20.8 / 100
Sources and data date
| Type | Evidence and data date |
|---|---|
| Price | Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. |
| Input | Documented modalities. Retrieved September 8, 2026. |
| Output | Documented modalities. Retrieved September 8, 2026. |
| Research status | Research date September 8, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | active-parameters-billions Original wording Active parameter count is not separately published for this dense model. |
| Not found in the inspected sources | api-model-id Original wording Gemma is distributed as open weights; the card does not define a hosted Gemini API model ID. |
| Additional source | Model metadata · Gemma 3 12B instruction-tuned weights · Editorial description reviewed October 2, 2026 |
| Additional source | Artificial Analysis (retrieved September 8, 2026) |
| Additional source | Model metadata (retrieved September 8, 2026) |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Creative Commons Attribution 4.0 |