Model: Gemma 4 12B IT

Google released Gemma 4 12B IT in June, a dense 12B-parameter model with a 256K context window that accepts text, images and audio directly in a single decoder without separate encoders. Only 8 of the model's 48 layers cache full context, each with a single wide KV head, keeping the KV cache to about 1 GB at 32K and 4.5 GB at the full 256K; at Q4_K_M (unsloth's 7.1 GB file) that works out to about 8.5 GB at 32K and 10.2 GB at 128K, both fitting on a 12 GB card, while an 8 GB card is tight even at 4K. The image and audio embedders ship as a separate mmproj file, and no per-level quality loss has been published.

Gemma 4 12B IT 12B Google Gemma 4 Google's mid-sized Gemma 4, released in June: a dense 12B with a 256K window. It takes text, images and audio without separate encoders — the raw inputs go straight into the one decoder. Only 8 of its 48 layers cache the full context, each with a single wide KV head, so the cache stays small: about 1 GB at 32K and 4.5 GB at the full 256K. At Q4 K M unsloth's 7.1 GB file that is about 8.5 GB at 32K and 10.2 GB at 128K, both comfortable on a 12 GB card; a 16 GB card holds the whole window comfortably, and an 8 GB card is tight even at 4K. bartowski's Q4 K M is a larger file 7.7 GB , so add about half a gigabyte if you use it. Sizes come from the published GGUF files; the image and audio embedders ship as a separate mmproj file, and no per-level quality loss has been published. 262K Max Context 2 Quant Variants GGUF Q8 0 Best Quality — Accuracy Retained Quantization Variants Per-quant VRAM, quality loss, and inference speed on RTX 4090 Measured = run on this site · Estimated = not run here: calculated, or a figure this site has not verified · Community = public reports | Format | Level | BPW | VRAM | PPL Loss | Speed | Source | Actions | |---|---|---|---|---|---|---|---| | GGUF | Q4 K M | 4.78 | 7.1 GB | — | — | Estimated | | | GGUF | Q8 0 | 8.51 | 12.7 GB | — | — | Estimated | | Which file each size is for A file size is not a memory figure: the VRAM column adds the KV cache at 4K context and runtime buffers to the weights. A vision projector mmproj , where the repo ships one, is a separate download and is in neither number. - GGUF Q4 K M unsloth/gemma-4-12b-it-GGUF https://huggingface.co/unsloth/gemma-4-12b-it-GGUF · file name not recorded · 7.12 GB 7,121,861,440 bytes · listing read 2026-10-20 - GGUF Q8 0 unsloth/gemma-4-12b-it-GGUF https://huggingface.co/unsloth/gemma-4-12b-it-GGUF · gemma-4-12b-it-Q8 0.gguf · 12.67 GB 12,669,647,680 bytes · listing read 2026-10-20 Similar models Compare with Gemma 4 https://quantized.uk/tools/compare/?a=gemma-4-12b&b=gemma-4-26b-a4b 26B-A4BConsumer GPUPro GPU15.3 GBmin VRAM—accuracy https://quantized.uk/quant-hub/gemma-4-26b-a4b/ Gemma 4 26B-A4B IT Google Gemma 4 Google's Gemma 4 mixture-of-experts model: about 25B parameters in the text model, roughly 4B active per token, a 256K window and image input. Only 5 of its 30 layers keep a cache that grows with context, and those use fewer, wider heads, so the cache stays small — about 0.9 GB at 32K and 5.3 GB at the full 256K. At Q4 K M that is about 18.8 GB at 32K: comfortable on a 24 GB card, which can still load the whole window tight, about 23.6 GB . A 16 GB card cannot hold it, since the Q4 K M file alone is 17.0 GB. The vision encoder is a separate 1.2 GB file on top. Sizes come from bartowski's GGUF files, whose Q4 K M works out at 5.40 bits per weight, well above the generic 4-bit rate; Unsloth's UD-Q4 K M is a different file, 0.09 GB smaller. No per-level quality loss has been published. 31BConsumer GPUPro GPU18.6 GBmin VRAM—accuracy https://quantized.uk/quant-hub/gemma-4-31b/ Gemma 4 31B IT Google Gemma 4 Google's dense Gemma 4 flagship: about 31B parameters, a 256K window and image input. Only 10 of its 60 layers cache the full context, and those use four wide KV heads, so 128K of context adds about 11 GB of cache where a conventional 60-layer design would need about 120 GB. At Q4 it is about 21 GB at 4K — on a 24 GB card that is right at the edge of comfortable, and tight from 32K. A 32 GB card holds 32K comfortably. Google also publishes its own quantization-aware QAT Q4 0 build, a single 17.7 GB file: about 20.1 GB at 4K, comfortable on 24 GB up to 16K. Its 1.2 GB vision encoder is a separate file, and with it loaded even 4K is tight on 24 GB. The Q4 K M and Q8 0 sizes use the calculator's generic rates, and no per-level quality loss has been published. E4BConsumer GPUMac / Apple Silicon4.6 GBmin VRAM—accuracy https://quantized.uk/quant-hub/gemma-4-e4b/ Gemma 4 E4B IT Google Gemma 4 Google's on-device Gemma 4: about 4.7B parameters doing the work plus a 2.8B per-layer embedding table, a 128K window, and image and audio input. Only 4 of its 42 layers keep a cache that grows with context, so the full 128K window costs about 2.0 GB of cache. At Q4 it is about 4.9 GB at 4K and 7.0 GB at the full window — an 8 GB card runs it comfortably at everyday context lengths. Sizes here count the whole model in GPU memory, which is what a Mac or a full GPU load uses; llama.cpp keeps the embedding table in system RAM instead, so on a discrete card it needs roughly 1.6 GB less VRAM than shown at Q4. Sizes use the calculator's generic rates — no GGUF file size could be checked. E2BConsumer GPUMac / Apple Silicon2.8 GBmin VRAM—accuracy https://quantized.uk/quant-hub/gemma-4-e2b/ Gemma 4 E2B IT Google Gemma 4 The smallest Gemma 4: about 2.3B parameters doing the work plus a 2.3B per-layer embedding table, a 128K window, and image and audio input. Its cache barely grows — 3 of 35 layers track the full context, about 0.8 GB at 128K. At Q4 it is about 3.0 GB at 4K and 3.8 GB at the full window — comfortable on any card in this index and on an 8 GB Mac. Sizes count the whole model in GPU memory; llama.cpp keeps the embedding table in system RAM, so on a discrete card it needs roughly 1.3 GB less VRAM than shown at Q4. Sizes use the calculator's generic rates — no GGUF file size could be checked. Running Gemma 4 12B IT locally At Q4 K M and 4K of context, Gemma 4 12B IT needs about 8.0 GB — 6.8 GB of weights, 0.53 GB of KV cache and a 0.7 GB activation buffer. The smallest cards in this index that clear that comfortably are 10 GB ones — 2 of them, from the Arc B570 10G up — and 67 of the 78 cards here do. These are calculated figures, not measurements: the estimate stops counting a card as comfortable at 88% of its VRAM, which is roughly the room a desktop session needs. What longer context costs Going from 4K to 32K adds about 0.44 GB, taking the total to 8.5 GB. That is gentler than the parameter count suggests: only 8 of 48 layers keep a KV cache that grows with context, and 40 more cap theirs at a sliding window, so the cache scales at a fraction of the usual rate. The model's native window is 256K; holding all of it at Q4 K M would need about 12 GB. Which build to download This index only tracks Gemma 4 12B IT in GGUF, across 2 levels Q4 K M, Q8 0 . No per-level perplexity sweep has been published for this model, so the quality column is empty rather than estimated — within one model, more bits per weight is the only ordering the data supports. No throughput has been measured for this model on this site's hardware. GGUF runs on llama.cpp and Ollama across NVIDIA, AMD and Apple silicon; AWQ and GPTQ target vLLM on CUDA and ROCm; EXL2 is ExLlamaV2 — CUDA only, and now archived, so local use rather than serving. Common questions - How much VRAM does Gemma 4 12B IT need? - About 8.0 GB at Q4 K M with 4K of context and batch 1, rising to roughly 8.5 GB at 32K. That figure is weights plus KV cache plus a 10% activation buffer, calculated from the model's architecture rather than measured on a card. - Will Gemma 4 12B IT run on a 10GB GPU? - Yes — at Q4 K M and 4K context it needs about 8.0 GB, which leaves 2.0 GB spare on a Arc B570 10G, or on any of the other 10 GB cards here 2 in all . That is the smallest size in this index that clears it comfortably; 67 of 78 do. - Which quantization of Gemma 4 12B IT should I use? - No published quality comparison exists for this model, so pick by footprint: Q4 K M is the level most people run, and a higher bits-per-weight level is more faithful. All 2 levels in the index are Q4 K M, Q8 0. Where this model fits Sized at Q4 K M with a 4K context window, smallest card first. Comfortable means the estimate uses at most 88% of the memory. - Runs comfortably on - Arc B570 10G10GB · +2.0 https://quantized.uk/gpu/arc-b570-10g/ RTX 3080 10G10GB · +2.0 https://quantized.uk/gpu/rtx-3080-10g/ RTX 3060 12G12GB · +4.0 https://quantized.uk/gpu/rtx-3060-12g/ Radeon RX 6700 XT12GB · +4.0 https://quantized.uk/gpu/radeon-rx-6700-xt/ Radeon RX 7700 XT12GB · +4.0 https://quantized.uk/gpu/radeon-rx-7700-xt/ - Tight but possible - RTX 40608GB https://quantized.uk/gpu/rtx-4060/ RTX 4060 Ti 8G8GB https://quantized.uk/gpu/rtx-4060-ti-8g/ Radeon RX 9060 XT 8G8GB https://quantized.uk/gpu/radeon-rx-9060-xt-8g/ - Just misses - Mac M1 8G+2.0GB over https://quantized.uk/gpu/mac-m1-8g/ Mac M3 8G+2.0GB over https://quantized.uk/gpu/mac-m3-8g/ - Ships in - GGUF https://quantized.uk/formats/gguf/ - Try it yourself - Size it in the calculator https://quantized.uk/tools/vram-calc/?model=gemma-4-12b&quant=Q4 K M Generate the run command https://quantized.uk/tools/cli-gen/?model=gemma-4-12b&quant=Q4 K M How to actually run this Deployment guides for this model and this class of hardware. This page's figures change when the model or the runtime does.Last updated 2026-10-23 RSS → /feed.xml https://quantized.uk/feed.xml