Updated minutes ago
ReleasedApril 9, 2026Verified 3d ago · ai.google.dev
Google

Gemini 3 Pro Preview

Google DeepMind · moe · 600B parameters · 1,000,000 context

Quality
50.0
Deploy this model →Use via API →Calculate ROI →Compare with othersFine-tune this model

Parameters

600B

Context Window

977K tokens

Architecture

MoE

Best GPU

B200 SXM

Cheapest API

$12.00/M

Intelligence Brief

Gemini 3 Pro Preview is a 600B parameter Mixture-of-Experts (16 experts, 2 active) model from Google DeepMind, featuring Grouped Query Attention (GQA) with 80 layers and 10,240 hidden dimensions. With a 1,000,000 token context window, it supports tools, vision, structured output, code, math, multilingual, reasoning. The most cost-effective API deployment is via google at $12.00/M output tokens. For self-hosted inference, B200 SXM delivers optimal throughput at $34088/month.

Provider pricing

1 provider · canonical: google
Provider Input $/M Output $/M ▲Notes
googlecanonical$2.00$12.00cheapest input · cheapest output

Prices update via the nightly pricing cron + admin approvals at /admin/ingest-queue. The leaderboard's Input/Output cells show the canonical rate above; this table shows the full spread.

Recent changes

Loading…

Related models

5 suggestions

Picks: same family first, then same vendor within ±2× params, then top tag-overlap matches. Price shown is the cheapest Output $/M across providers — the row's page shows the canonical anchor.

Architecture Details

TypeMOE
Total Parameters600B
Active Parameters100B
Layers80
Hidden Dimension10,240
Attention Heads80
KV Heads10
Head Dimension128
Vocab Size200,000
Total Experts16
Active Experts2

Memory Requirements

BF16 Weights

1200.0 GB

FP8 Weights

600.0 GB

INT4 Weights

300.0 GB

KV-Cache per Token204800 bytes
Activation Estimate4.00 GB

Fits on (single GPU) — most practical first

BEST FITB200 NVL (pair)360GBINT4

Fits on (multi-GPU with Tensor Parallelism)

RECOMMENDEDx2B200 SXM180GB eachINT4x2H100 NVL 94GB (per GPU pair)188GB eachINT4x2B100 SXM192GB eachINT4x2GB200 NVL72 (per GPU)192GB eachINT4x2GB300 NVL72 (per GPU)192GB eachINT4x2Instinct MI300X192GB eachINT4x2Groq LPU230GB eachINT4x2Instinct MI325X256GB eachINT4x2B300288GB eachINT4x3Instinct MI250X128GB eachINT4x3Gaudi 3128GB eachINT4

Multi-GPU configurations use Tensor Parallelism (TP) to split model layers across GPUs. Requires NVLink or NVSwitch interconnect for optimal performance.

GPU Compatibility Matrix

Gemini 3 Pro Preview is compatible with 1% of GPU configurations across 41 GPUs at 3 precision levels.

BF16 (Full)
FP8 (Half)
INT4 (Quarter)
Blackwell(7 GPUs)
B200 NVL (pair)360GB
✗
✗
17%
B300288GB
✗
✗
✗
B100 SXM192GB
✗
✗
✗
GB200 NVL72 (per GPU)192GB
✗
✗
✗
Hopper(7 GPUs)
H100 NVL 94GB (per GPU pair)188GB
✗
✗
✗
H200 SXM141GB
✗
✗
✗
H2096GB
✗
✗
✗
GH20096GB
✗
✗
✗
Ada Lovelace(11 GPUs)
L40S48GB
✗
✗
✗
L4048GB
✗
✗
✗
RTX 6000 Ada48GB
✗
✗
✗
L2048GB
✗
✗
✗
Ampere(16 GPUs)
A100 80GB SXM80GB
✗
✗
✗
A100 80GB PCIe80GB
✗
✗
✗
A1664GB
✗
✗
✗
RTX A600048GB
✗
✗
✗
Legend:No fitVery tightTightModerateGoodExcellent

GPU Recommendations

B200 SXMgood

BF16 · 8 GPUs · tensorrt-llm

73/100

score

Throughput

140.0 tok/s

Latency (ITL)

7.1ms

Est. TTFT

1ms

Cost/Month

$34088

Cost/M Tokens

$92.65

Use this config →
B100 SXMgood

BF16 · 8 GPUs · tensorrt-llm

73/100

score

Throughput

140.0 tok/s

Latency (ITL)

7.1ms

Est. TTFT

1ms

Cost/Month

$34164

Cost/M Tokens

$92.86

Use this config →
H200 SXMgood

BF16 · 16 GPUs · tensorrt-llm

70/100

score

Throughput

140.0 tok/s

Latency (ITL)

7.1ms

Est. TTFT

1ms

Cost/Month

$40845

Cost/M Tokens

$111.02

Use this config →

Deployment Options

API

API Deployment

google

$12.00/M

output tokens

Self-Hosted

Single GPU

Requires multi-GPU setup (600 GB VRAM needed)

Scale

Multi-GPU

B200 SXM x8

140.0 tok/s

TP· $34088/mo

Open Full Calculator →

API Pricing Comparison

ProviderInput $/MOutput $/MBadges
google$2.00$12.00
Cheapest

Cost Analysis

ProviderInput $/MOutput $/M~Monthly Cost
googleBest Value$2.00$12.00$70

Cost per 1,000 Requests

Short (500 tok)

$3.40

via google

Medium (2K tok)

$13.60

via google

Long (8K tok)

$40.00

via google

Performance Estimates

Throughput by GPU

B200 SXM
140.0 tok/s
B100 SXM
140.0 tok/s
H200 SXM
140.0 tok/s

VRAM Breakdown (B200 SXM, BF16)

Weights
Act
Weights 150.0 GBKV-Cache 6.7 GBActivations 32.0 GBOverhead 7.5 GB

Capabilities

Features

✓ Tool Use✓ Vision✓ Code✓ Math✓ Reasoning✓ Multilingual✓ Structured Output

Supported Frameworks

vllm

Supported Precisions

BF16 (default)

Where to Deploy Gemini 3 Pro Preview

Similar Models

Gemini 2.0 Pro

600B params · moe

Quality: 88

from $4.00/M

Larger context, Higher quality, CheaperCompare →

Grok 3

600B params · moe

Quality: 90

from $15.00/M

Smaller context, Higher qualityCompare →

Megatron-Turing NLG 530B

530B params · dense

Quality: 58

Smaller context, Higher qualityCompare →

DeepSeek R1

671B params · moe

Quality: 88

from $2.00/M

Smaller context, Higher quality, CheaperCompare →

DeepSeek V3

671B params · moe

Quality: 81

from $0.42/M

Smaller context, Higher quality, CheaperCompare →

Frequently Asked Questions

How much VRAM does Gemini 3 Pro Preview need for inference?

Gemini 3 Pro Preview requires approximately 1200.0 GB of VRAM at BF16 precision, 600.0 GB at FP8, or 300.0 GB at INT4 quantization. Additional VRAM is needed for KV-cache (204800 bytes per token) and activations (~4.00 GB).

What is the best GPU for Gemini 3 Pro Preview?

The top recommended GPU for Gemini 3 Pro Preview is the B200 SXM (x8) using BF16 precision. It achieves approximately 140.0 tokens/sec at an estimated cost of $34088/month ($92.65/M tokens). Score: 73/100.

How much does Gemini 3 Pro Preview inference cost?

Gemini 3 Pro Preview API inference starts from $2.00/M input tokens and $12.00/M output tokens. Self-hosted inference costs depend on your GPU configuration — use our ROI calculator for a detailed breakdown.