standardGemma Terms of Use12B paramsOpen weights

Gemma 3 12B

12B parameter open-weight model. 128K context at 12B. Q4 fits in 8–10 GB VRAM with short context; KV cache grows quickly at 128K. Benchmark against Qwen3 14B on your tasks.

Google · Gemma 3

Model overview

Gemma 3 12B is Google's mid-size Gemma 3 checkpoint, built for chat, RAG, and summarization tasks that benefit from its 128K context window without needing a 30B+ model. At Q4 it fits comfortably in an 8-10 GB consumer GPU for short-to-medium prompts, though KV cache usage grows quickly as you push toward the full 128K window, so budget extra VRAM headroom for long-context sessions. It ships under Google's Gemma Terms of Use rather than a fully permissive open-source license, so review the terms before commercial use. It's worth benchmarking against similarly sized Qwen3 checkpoints, since relative strengths vary by task.

Check your GPU fitWhat is VRAM?What is quantization?Local LLM quickstart

Editorial review

Reviewed byOpenSourcesAI EditorialLast updatedJune 2026SourcesHuggingFace model card (google/gemma-3-12b-it), official docs, OpenSourcesAI editorial review.

VRAM figures are empirical estimates. Actual usage varies by runtime, context length, and system configuration. Verify on your specific hardware before production use.

Will Gemma 3 12B run on your machine?

Gemma 3 12B is 12B parameters and needs 9.1 GB of VRAM at Q4_K_M — 7.6 GB of weights plus 1.5 GB of runtime overhead for the inference server itself.

VRAM by quantization

QuantizationWeightsNeeds (with overhead)Quality
Q4_K_M7.6 GB9.1 GBgood
Q8_013 GB14.5 GBhigh
FP1624 GB25.5 GBreference

Fit on common hardware at Q4_K_M

HardwareMemory the model can useSystem RAMVerdict
CPU OnlyNone (CPU only)16 GBCPU offload
RTX 4060 Laptop8 GB16 GBCPU offload
RTX 3060 (12GB)12 GB32 GBComfortable
RTX 4060 Ti (16GB)16 GB32 GBComfortable
RTX 309024 GB64 GBComfortable

Comfortable means VRAM clears the requirement by 2 GB or more. Tight means it covers the requirement with no margin. CPU offload means the model does not fit in VRAM but system RAM is at least 1.6× the weights, so it will run at reduced speed — expect roughly 1–5 tokens per second. Figures are weights plus a fixed runtime overhead and exclude KV-cache growth, which scales with context length.

Check Gemma 3 12B against your exact hardwareHow these verdicts are calculated

Need more hardware for Gemma 3 12B? Open the PC Builder for the 13B / 14B tier →

Related models

Gemma 4 12B12B parameter model — shared focus: chat, reasoning.Gemma 4 12B (QAT)12B parameter model — shared focus: chat, coding, summarisation.DeepSeek-R1 Distill 14B14B parameter model — shared focus: reasoning, coding, agents.CodeGemma 7BCoding assistant experiments and developer workflow prototypes.

Ready to run this model locally?

Find a compatible interface in our Local AI Tools directory →