LLM Reference

Gemma 4 12B IT

Released
2026-06-03
Last refreshed
2026-06-03
Status
Researched today
Open SourceMultimodalCodingRAGAgentsLong contextVisionClassificationJSON / Tool useopen-sourcemultimodal

Gemma 4 12B IT is worth evaluating for coding, rag, and agents when its provider route and context window match the workload.

Use it for

  • Teams evaluating coding, rag, and agents
  • Workloads that can use a 256k context window
  • Buyers comparing 2 tracked provider routes

Do not use it for

  • Workloads where another current model has stronger sourced task evidence
Specifications
Family
Gemma 4
Released
2026-06-03
Context
256k
Parameters
11.9B
Architecture
encoder_free_unified_multimodal
Knowledge cutoff
2025-01
Specialization
general
License
Apache 2.0
Training
finetuned
Created by

Pioneering artificial intelligence research.

London, United Kingdom
Founded 2014
Website
Pricing
Output / 1M
-
Input / 1M
-

Cheapest of 2 routes · Hugging Face Inference Endpoints

About

Instruction-tuned 12B Gemma 4 model with native text, image, video, and audio input through an encoder-free unified architecture. It runs on 16 GB VRAM in BF16, supports a 256K context window, configurable thinking mode, function calling, structured outputs, and 140+ languages, making it the mid-sized Gemma 4 option between E4B and the 26B MoE.

Gemma 4 12B IT is an open-source model in the Gemma 4 family. The structured metadata tracks a 256k-token context window, multimodal input, audio, reasoning, function calling, tool use, and structured outputs. This page tracks provider routes through Hugging Face Inference Endpoints and Kaggle Models. Headline tracked benchmarks include Google-Proof Q&A 78.8, MMLU PRO 77.2, and LiveCodeBench 72.0.

Top use-case fit

Coding

1 relevant benchmark in the decision map.

RAG

Included by capability and metadata signals in the decision map.

Agents

Included by capability and metadata signals in the decision map.

Provider price ladder

Compare all 2
ProviderInput / 1MOutput / 1MRoute
Hugging Face Inference Endpoints--
Partial
Kaggle Models--
Partial

Capabilities

VisionMultimodalReasoningFunction CallingTool UseStructured OutputsAudio

Benchmark peer barsfor Coding

Benchmark scores(4)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
BenchmarkScoreVersionSource
Google-Proof Q&A78.8Diamondhttps://huggingface.co/google/gemma-4-12B-it
MMLU PRO77.2MMLU Prohttps://huggingface.co/google/gemma-4-12B-it
LiveCodeBench72.0v6 pass@1https://huggingface.co/google/gemma-4-12B-it
AIME 202677.5no tools / no calculatorhttps://huggingface.co/google/gemma-4-12B-it

Migration checks

No linked migration route is available for this model yet.

Rankings & picks(10)