Gemma 4 12B IT

Tool callingStructured outputReasoningImage input

Model card Launch post Website

Compact Gemma 4 instruction model for open, self-hosted chat and reasoning

Context: 262K
Released: 4mo ago
Parameters: 12B

Sources:models.dev

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

No benchmark results have been matched to this model yet.

Specifications

Developer: Google DeepMind
Model family: Gemma
Type: Language model
Predecessor: —
Successor: —
Released: June 9, 2026
Model ID: google/gemma-4-12b-it
Openness: Open weights
License: —
Parameters: 12B
Input modalities: text, image
Output modalities: text
Modality: text+image->text
Max output tokens: 33K
Hugging Face: —
Source freshness: Metadata synced Oct 8, 2026, 10:17 PM UTC

Additional source evidence

Source model: google/gemma-4-12b-it
Source family: gemma
Last updated: 2026-06-09
Model context: 262K
Model output limit: 33K
Open weights: Yes
File attachments: Yes
Reasoning: Yes
Tool calling: Yes
Structured output: Yes
Temperature control: Yes

Downloadable weights

Hugging Face

Complete source record (JSON)

{
  "id": "google/gemma-4-12b-it",
  "name": "Gemma 4 12B IT",
  "description": "Compact Gemma 4 instruction model for open, self-hosted chat and reasoning",
  "family": "gemma",
  "last_updated": "2026-06-09",
  "attachment": true,
  "reasoning": true,
  "tool_call": true,
  "structured_output": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text",
      "image"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 262144,
    "output": 32768
  },
  "release_date": "2026-06-09",
  "open_weights": true,
  "weights": [
    {
      "label": "Hugging Face",
      "url": "https://huggingface.co/google/gemma-4-12B-it"
    }
  ]
}

Infrastructure

Provider availability

All providers

Provider endpoint data will appear after this model's first OpenRouter endpoint sync.

Text provenance

Gemini app/web

Google SynthID text watermarking

Google DeepMind says text generated in the Gemini app and web experience is watermarked with SynthID by adjusting token probabilities.

Official explanation

Deployment graph

Can I run this?

Vertical Tabs only shows compatibility when it has structured evidence tying this exact model to hardware, runtime, quantization, memory use, and observed performance.

Browse hardware