Models / Gemma 4 31B / Calculator

Gemma 4 31B cost & VRAM calculator

What does Gemma 4 31B cost to run for your workload, and can you run it on your own hardware? Set your workload below - we compute per-provider API cost live and tell you honestly whether local hardware can run it.

Your workload

Casual Power user Team of 5 Codebase task

Gemma 4 31B runs an always-on thinking mode. Reasoning (thinking) tokens are billed at the output rate ($15.00/M), so count them here to see the thinking portion of your bill.

API cost for your workload

Provider Rate ($/1M) Monthly cost
CoreWeave $0.10 in · $0.34 out · $0.10 cache $0.0 cheapest
Venice $0.12 in · $0.36 out · $0.09 cache $0.0 cheapest
Chutes $0.12 in · $0.37 out · $0.01 cache $0.0 cheapest
Novita $0.14 in · $0.40 out · $0.10 cache $0.0 cheapest
Crusoe $0.14 in · $0.40 out · $0.14 cache $0.0 cheapest
Friendli $0.14 in · $0.40 out · $0.10 cache $0.0 cheapest
Parasail $0.15 in · $0.40 out · $0.06 cache $0.0 cheapest
DeepInfra $0.30 in · $0.60 out · $0.10 cache $0.01
Io Net $0.36 in · $1.09 out · $0.18 cache $0.01
SambaNova $0.38 in · $1.15 out · $0.10 cache $0.01
Together AI may be stale $0.39 in · $0.97 out $0.01
SiliconFlow $0.75 in · $1.00 out · $0.25 cache $0.02
ModelRun $0.75 in · $1.00 out · $0.20 cache $0.02

Monthly cost is an estimate from list prices and your workload - verify against the provider before committing. Cached fraction applies the cache rate to that share of input.

Can you run it locally?

YES - IT RUNS ON HARDWARE YOU CAN BUY

Gemma 4 31B fits on individual rigs. Here's the hardware that runs it comfortably; the full per-quant VRAM matrix and tok/s estimates live on the model page.

4x H100 80GB (320GB) 320GB QAT
409.4t/s
4x RTX 5090 (128GB) 128GB NVFP4
246.4t/s
NVIDIA DGX Station 748GB 748GB QAT
244.4t/s
8x RTX 3090 rack (192GB) 192GB QAT
228.8t/s
AMD Instinct MI300X (192GB) 192GB QAT
162.7t/s
2x RTX 5090 (64GB) 64GB NVFP4
123.2t/s
4x RTX 4090 (96GB) 96GB QAT
123.2t/s
Single RTX 5090 (32GB) 32GB NVFP4
61.6t/s
RTX PRO 6000 Blackwell (96GB) 96GB NVFP4
61.6t/s
2x RTX 3090 (48GB) 48GB QAT
57.2t/s
Mac Studio M4 Ultra 192GB 192GB NVFP4
40.9t/s
Mac Studio M4 Ultra 512GB 512GB NVFP4
40.9t/s
Single RTX 4090 (24GB) 24GB QAT
30.8t/s
MacBook Pro M5 Max 128GB 128GB NVFP4
23.0t/s
Dual EPYC 9004 + 768GB DDR5-4800 768GB QAT
14.1t/s
DGX Spark 128GB unified 128GB NVFP4
9.4t/s
Ryzen AI Max+ 395 128GB 128GB QAT
7.8t/s
Jetson AGX Orin 64GB 64GB QAT
6.3t/s
Epyc + 512GB DDR4-3200 + 2x RTX 3090 560GB QAT
6.3t/s
Epyc + 512GB DDR4-2400 + 2x RTX 3090 560GB QAT
4.7t/s
QAT
18.0GB weights 19.0GB min 22.0GB rec
Q4_K_M
18.3GB weights 20.0GB min 22.0GB rec
NVFP4
16.0GB weights 20.0GB min 22.0GB rec
Q8_0
32.6GB weights 34.0GB min 38.0GB rec
BF16
61.4GB weights 63.0GB min 66.0GB rec

Want the own-vs-rent payback (capex vs your monthly API bill)? The budget tool computes that against representative rigs.

Full model card API pricing table Generic token calculator