Updated minutes ago
Anthropic

Claude Opus 4

Anthropic · dense · 200B parameters · 200,000 context

Quality
90.0
Calculate ROI →Compare with othersFine-Tune This Model →

Parameters

200B

Context Window

195K tokens

Architecture

Dense

Best GPU

B200 NVL (pair)

Cheapest API

$75.00/M

Quality Score

90/100

Intelligence Brief

Claude Opus 4 is a 200B parameter DENSE model from Anthropic, featuring Grouped Query Attention (GQA) with 96 layers and 12,288 hidden dimensions. With a 200,000 token context window, it supports tools, vision, structured output, code, math, multilingual, reasoning. On standardized benchmarks, it achieves MMLU 91.5, HumanEval 75, GSM8K 97. The most cost-effective API deployment is via anthropic at $75.00/M output tokens. For self-hosted inference, B200 NVL (pair) delivers optimal throughput at $19929/month.

Architecture Details

TypeDENSE
Total Parameters200B
Active Parameters200B
Layers96
Hidden Dimension12,288
Attention Heads96
KV Heads16
Head Dimension128
Vocab Size152,064

Memory Requirements

BF16 Weights

400.0 GB

FP8 Weights

200.0 GB

INT4 Weights

100.0 GB

KV-Cache per Token393216 bytes
Activation Estimate5.00 GB

Fits on (single GPU) — most practical first

BEST FITInstinct MI250X128GBINT4Gaudi 3128GBINT4Gaudi 3 HL-325L128GBINT4H200 SXM141GBINT4B200 SXM180GBINT4H100 NVL 94GB (per GPU pair)188GBINT4B100 SXM192GBINT4GB200 NVL72 (per GPU)192GBINT4GB300 NVL72 (per GPU)192GBINT4Instinct MI300X192GBINT4Groq LPU230GBINT4Instinct MI325X256GBFP8

GPU Compatibility Matrix

Claude Opus 4 is compatible with 8% of GPU configurations across 41 GPUs at 3 precision levels.

BF16 (Full)
FP8 (Half)
INT4 (Quarter)
Blackwell(7 GPUs)
B200 NVL (pair)360GB
✗
44%
72%
B300288GB
✗
31%
65%
B100 SXM192GB
✗
✗
48%
GB200 NVL72 (per GPU)192GB
✗
✗
48%
Hopper(7 GPUs)
H100 NVL 94GB (per GPU pair)188GB
✗
✗
47%
H200 SXM141GB
✗
✗
29%
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 NVL (pair)optimal

BF16 · 2 GPUs · tensorrt-llm

83/100

score

Throughput

280.0 tok/s

Latency (ITL)

3.6ms

Est. TTFT

1ms

Cost/Month

$19929

Cost/M Tokens

$27.08

Use this config →
H20optimal

BF16 · 8 GPUs · tensorrt-llm

80/100

score

Throughput

280.0 tok/s

Latency (ITL)

3.6ms

Est. TTFT

1ms

Cost/Month

$7516

Cost/M Tokens

$10.21

Use this config →
B200 SXMgood

BF16 · 4 GPUs · tensorrt-llm

78/100

score

Throughput

280.0 tok/s

Latency (ITL)

3.6ms

Est. TTFT

1ms

Cost/Month

$17044

Cost/M Tokens

$23.16

Use this config →

Deployment Options

API

API Deployment

anthropic

$75.00/M

output tokens

Self-Hosted

Single GPU

Requires multi-GPU setup (200 GB VRAM needed)

Scale

Multi-GPU

B200 NVL (pair) x2

280.0 tok/s

TP· $19929/mo

Open Full Calculator →

API Pricing Comparison

ProviderInput $/MOutput $/MBadges
anthropic$15.00$75.00
Cheapest

Cost Analysis

ProviderInput $/MOutput $/M~Monthly Cost
anthropicBest Value$15.00$75.00$450

Cost per 1,000 Requests

Short (500 tok)

$22.50

via anthropic

Medium (2K tok)

$90.00

via anthropic

Long (8K tok)

$270.00

via anthropic

Performance Estimates

Throughput by GPU

B200 NVL (pair)
280.0 tok/s
H20
280.0 tok/s
B200 SXM
280.0 tok/s

VRAM Breakdown (B200 NVL (pair), BF16)

Weights
Act
Weights 200.0 GBKV-Cache 12.9 GBActivations 40.0 GBOverhead 10.0 GB

Quality Benchmarks

Top 10%
98th percentile across all models
MMLU
91.5
Top 10% (97th pctile)
HumanEval
75.0
Top 10% (93th pctile)
GSM8K
97.0
Top 10% (94th pctile)
MT-Bench
92.0
Bottom 25% (0th pctile)

Capabilities

Features

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

Supported Frameworks

Supported Precisions

BF16 (default)

Where to Deploy Claude Opus 4

Similar Models

Claude 3 Opus

175B params · dense

Quality: 80

from $75.00/M

Lower qualityCompare →

Claude Sonnet 4

70B params · dense

Quality: 86

from $15.00/M

Cheaper, Smaller modelCompare →

Claude 3 Sonnet

70B params · dense

Quality: 78

from $15.00/M

Lower quality, Cheaper, Smaller modelCompare →

GPT-4o

200B params · moe

Quality: 85

from $10.00/M

CheaperCompare →

GPT-4 Turbo

200B params · moe

Quality: 80

from $30.00/M

Lower quality, CheaperCompare →

Frequently Asked Questions

How much VRAM does Claude Opus 4 need for inference?

Claude Opus 4 requires approximately 400.0 GB of VRAM at BF16 precision, 200.0 GB at FP8, or 100.0 GB at INT4 quantization. Additional VRAM is needed for KV-cache (393216 bytes per token) and activations (~5.00 GB).

What is the best GPU for Claude Opus 4?

The top recommended GPU for Claude Opus 4 is the B200 NVL (pair) (x2) using BF16 precision. It achieves approximately 280.0 tokens/sec at an estimated cost of $19929/month ($27.08/M tokens). Score: 83/100.

How much does Claude Opus 4 inference cost?

Claude Opus 4 API inference starts from $15.00/M input tokens and $75.00/M output tokens. Self-hosted inference costs depend on your GPU configuration — use our ROI calculator for a detailed breakdown.