Models / Gemma 3 12B

Gemma 3 12B

consumer
HuggingFace → Google → Find for my hardware

AI-generated content marks

The provider reports that this model does not add embedded watermarks or provenance metadata to generated output.

Provider transparency docs →

coding reasoning chat
Parameters
12.2B
Context
128k
License
other
Developer
Google
Origin
🇺🇸 USA
Released
Feb 2025

Scores

Coding
62
Reasoning
65
General
68

Score per dollar

1360 pts per $/M input

general_score (68) divided by cheapest input price ($0.05/M). Higher is better value. See live pricing.

Related models

Gemma 4 26B A4B
25.2B 3.8B active enthusiast 15GB min RAM $0.04/M in Apr 30, 2026
Gemma 4 31B
30.7B enthusiast 19GB min RAM $0.08/M in Apr 2, 2026
Gemma 3 270M
270M edge 1GB min RAM Aug 14, 2025
Gemma 3n E2B
2.0B edge 6GB min RAM Jun 26, 2025
Gemma 3n E4B
4.0B edge 8GB min RAM Jun 26, 2025
Gemma 4 12B
12.0B consumer 7GB min RAM Jun 1, 2025
Gemma 3 1B
1.0B edge 1GB min RAM Feb 27, 2025
Gemma 3 27B
27.2B consumer 17GB min RAM $0.08/M in Feb 27, 2025
Gemma 3 4B
4.3B edge 3GB min RAM $0.05/M in Feb 27, 2025

Save your hardware and every model page answers the real question: will it run on your machine, and how fast?

Join free - save your rig →

Run it locally

Per-quant memory needs and a static "can you run it?" reference - no rig entry required

Q4_K_M
7.5GB 7.5GB min 8.5GB rec
Balanced - the usual local sweet spot
Q8_0
13.5GB 13.5GB min 15.0GB rec
Near-lossless

The reference hardware

Schematic of the 4x H100 80GB (320GB) reference rig - 320GB VRAM, 13400 GB/s aggregate bandwidth
Schematic of the NVIDIA DGX Station 748GB reference rig - 748GB unified memory, 8000 GB/s aggregate bandwidth
Schematic of the 8x RTX 3090 rack (192GB) reference rig - 192GB VRAM, 7489 GB/s aggregate bandwidth
Schematic of the 4x RTX 5090 (128GB) reference rig - 128GB VRAM, 7168 GB/s aggregate bandwidth
Schematic of the AMD Instinct MI300X (192GB) reference rig - 192GB VRAM, 5324 GB/s aggregate bandwidth
Schematic of the 4x RTX 4090 (96GB) reference rig - 96GB VRAM, 4032 GB/s aggregate bandwidth
Schematic of the 2x RTX 5090 (64GB) reference rig - 64GB VRAM, 3584 GB/s aggregate bandwidth
Schematic of the 2x RTX 3090 (48GB) reference rig - 48GB VRAM, 1872 GB/s aggregate bandwidth
Schematic of the Single RTX 5090 (32GB) reference rig - 32GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the RTX PRO 6000 Blackwell (96GB) reference rig - 96GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 192GB reference rig - 192GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 512GB reference rig - 512GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the Single RTX 4090 (24GB) reference rig - 24GB VRAM, 1008 GB/s aggregate bandwidth
Schematic of the MacBook Pro M5 Max 128GB reference rig - 128GB unified memory, 614 GB/s aggregate bandwidth
Schematic of the Single GTX 1080 Ti (11GB) reference rig - 11GB VRAM, 484 GB/s aggregate bandwidth
Schematic of the Dual EPYC 9004 + 768GB DDR5-4800 reference rig - 768GB unified memory, 460 GB/s aggregate bandwidth
Schematic of the DGX Spark 128GB unified reference rig - 128GB unified memory, 273 GB/s aggregate bandwidth
Schematic of the Ryzen AI Max+ 395 128GB reference rig - 128GB unified memory, 256 GB/s aggregate bandwidth
Schematic of the Jetson AGX Orin 64GB reference rig - 64GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-3200 + 2x RTX 3090 reference rig - 560GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-2400 + 2x RTX 3090 reference rig - 560GB unified memory, 153 GB/s aggregate bandwidth
Schematic of the NVIDIA Jetson Orin NX 16GB reference rig - 16GB unified memory, 102 GB/s aggregate bandwidth

22 reference configs, drawn in-house. Scroll for more.

Can you run it? - reference rigs

Rig Q4_K_M Q8_0
4x H100 80GB (320GB) fast 982.5t/s fast 545.9t/s
NVIDIA DGX Station 748GB fast 586.5t/s fast 325.9t/s
8x RTX 3090 rack (192GB) fast 549.1t/s fast 305.1t/s
4x RTX 5090 (128GB) fast 525.5t/s fast 292.0t/s
AMD Instinct MI300X (192GB) fast 390.4t/s fast 216.9t/s
4x RTX 4090 (96GB) fast 295.6t/s fast 164.3t/s
2x RTX 5090 (64GB) fast 262.8t/s fast 146.0t/s
2x RTX 3090 (48GB) fast 137.3t/s fast 76.3t/s
Single RTX 5090 (32GB) fast 131.4t/s fast 73.0t/s
RTX PRO 6000 Blackwell (96GB) fast 131.4t/s fast 73.0t/s
Mac Studio M4 Ultra 192GB fast 87.3t/s fast 48.5t/s
Mac Studio M4 Ultra 512GB fast 87.3t/s fast 48.5t/s
Single RTX 4090 (24GB) fast 73.9t/s fast 41.1t/s
MacBook Pro M5 Max 128GB fast 49.1t/s fast 27.3t/s
Single GTX 1080 Ti (11GB) fast 35.5t/s offload
Dual EPYC 9004 + 768GB DDR5-4800 fast 33.8t/s ok 18.8t/s
DGX Spark 128GB unified fast 20.0t/s ok 11.1t/s
Ryzen AI Max+ 395 128GB ok 18.8t/s ok 10.4t/s
Jetson AGX Orin 64GB ok 15.0t/s ok 8.3t/s
Epyc + 512GB DDR4-3200 + 2x RTX 3090 ok 15.0t/s ok 8.3t/s
Epyc + 512GB DDR4-2400 + 2x RTX 3090 ok 11.3t/s slow 6.3t/s
NVIDIA Jetson Orin NX 16GB slow 7.5t/s slow 4.2t/s

Fit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.

Download options

Q4_K_M official -5% vs fp16
7.5GB dl 7.5GB min 8.5GB rec
REC RAM vs largest quant
Q8_0 official -1% vs fp16
13.5GB dl 13.5GB min 15.0GB rec
REC RAM vs largest quant

Or run it in the cloud

Live per-provider pricing, throughput and uptime - refreshed about 9 hours ago via OpenRouter. Click a column to sort.

Provider Type Input $/M Output $/M Cache $/M Tok/s Latency Uptime Value
DeepInfra
API 0.05 0.15 - - - 100.00% best uptime

Default order: throughput among 95%+ uptime providers, then latency; subscriptions last. Sort by any column. Subscription rows show $/mo in the Value column - per-token columns are "-". Affiliate links are marked sponsored / nofollow. Confirm current pricing on the provider's site before committing.

Detailed API pricing page + JSON endpoint →

See who runs Google in production →

PRICE HISTORY

Inference cost over time

Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.

Loading price history...