Anthropic Modelle

Entdecken Sie alle 6 Modelle von Anthropic mit detaillierten Preisen, Vor- und Nachteilen sowie Entwicklerempfehlungen.

Modelle

$0.800

Niedrigster Input

Max. Kontext

Qualitätsstufen

Schnellempfehlungen

Bestes Preis-Leistungs-Verhältnis: Claude 3.5 Haiku ($0.800/1M)

Beste Qualität: Claude Opus 4.8

Claude Opus 4.8

Flagship

Frontier coding, AI agents

Offizielle Preise

Wann verwenden: When you need the absolute latest Anthropic intelligence for complex coding and agent workflows.

Upgrade-Highlights

◆Latest Anthropic flagship: knowledge cutoff 2025-06 (2 months newer than 4.7)
◆Fast mode available for premium speed — 2x faster responses
◆1M context window (up from 200K in Claude 4 Opus)
◆Same $5/$25 pricing as Opus 4.7 — more capability at same cost
◆Improved agentic coding: better multi-step tool orchestration

Input-Preis

$5.00

per 1M tokens

Output-Preis

$25.00

per 1M tokens

Cached Input

$0.500

per 1M tokens

Batch-Input

—

per 1M tokens

Kontextfenster: 1M

Max. Output: 32,000 tokens

Wissensstand: 2025-06

VisionFunktionsaufrufFeinabstimmungJSON-ModusKostenlose Stufe

Vorteile

Latest Anthropic flagship with 1M context
Fast mode available for premium speed
Best-in-class agentic coding

Nachteile

Expensive output at $25/M tokens
No batch API
No fine-tuning

Leistung

Ausgabegeschwindigkeit~65 tok/s

Rate-Limit4,000 RPM

Multimodal

BildeingabeBildausgabeAudioeingabeAudioausgabe

Benchmarks

SWE-bench Verified

79.8%

MMLU

89.8%

GPQA

78.5%

Claude Opus 4.7

Flagship

Agentic coding, complex analysis

Offizielle Preise

Wann verwenden: Top choice for long-horizon coding agents and complex analytical workloads.

Upgrade-Highlights

◆SWE-Bench Pro: 64.3% — leads all models on real-world coding
◆1M context window — 5x increase over Claude 4 Opus's 200K
◆Same $5/$25 pricing as Opus 4.6 with better performance
◆Knowledge cutoff: 2025-04 — 1 month newer than Claude 4
◆Best-in-class for long-horizon multi-step coding agents

Input-Preis

$5.00

per 1M tokens

Output-Preis

$25.00

per 1M tokens

Cached Input

$0.500

per 1M tokens

Batch-Input

—

per 1M tokens

Kontextfenster: 1M

Max. Output: 32,000 tokens

Wissensstand: 2025-04

VisionFunktionsaufrufFeinabstimmungJSON-ModusKostenlose Stufe

Vorteile

1M context window
64.3% on SWE-Bench Pro (leads competition)
Same price as Opus 4.6 with better performance

Nachteile

Still expensive at $5/$25
No batch API
No fine-tuning

Leistung

Ausgabegeschwindigkeit~60 tok/s

Rate-Limit4,000 RPM

Multimodal

BildeingabeBildausgabeAudioeingabeAudioausgabe

Benchmarks

SWE-bench Pro

64.3%

MMLU

89.5%

GPQA

77.0%

Agenten die dieses Modell verwenden

Anthropic Computer Use

Claude Haiku 4.5

Lite

Fast, cost-efficient AI tasks

Offizielle Preise

Wann verwenden: Best Anthropic value for production chatbots, coding assistants, and agent tasks.

Upgrade-Highlights

◆Matches Sonnet 4 coding performance at 1/3 the price
◆Vision support at lite tier — first Haiku with multimodal capability
◆200K context window (same as Claude 4 Opus/Sonnet)
◆Knowledge cutoff: 2025-02 — newest at the Haiku tier
◆Price: $1/$5 — 5x more than 3.5 Haiku but 3x quality improvement

Input-Preis

$1.00

per 1M tokens

Output-Preis

$5.00

per 1M tokens

Cached Input

$0.100

per 1M tokens

Batch-Input

—

per 1M tokens

Kontextfenster: 200K

Max. Output: 8,192 tokens

Wissensstand: 2025-02

VisionFunktionsaufrufFeinabstimmungJSON-ModusKostenlose Stufe

Vorteile

Matches Sonnet 4 performance on coding tasks
Vision support at lite tier
200K context window

Nachteile

5x more expensive than 3.5 Haiku
8K max output is limiting
No fine-tuning

Leistung

Ausgabegeschwindigkeit~100 tok/s

Rate-Limit8,000 RPM

Multimodal

BildeingabeBildausgabeAudioeingabeAudioausgabe

Benchmarks

SWE-bench Verified

58.0%

MMLU

85.0%

HumanEval

84.5%

Claude 4 Opus

Flagship

Agentic coding, deep analysis

Offizielle Preise

Wann verwenden: When you need the absolute best for complex agentic coding and deep analytical work.

Upgrade-Highlights

◆SWE-bench Verified: 72.7% — highest agentic coding score at launch
◆200K context with 32K max output — 4x Claude 3.5 Sonnet's 8K output
◆Batch API available: $7.50/$37.50 — 50% savings for async workloads
◆Knowledge cutoff: 2025-03 — 6 months newer than 3.5 Sonnet
◆Premium pricing ($15/$75) reflects highest intelligence tier

Input-Preis

$15.00

per 1M tokens

Output-Preis

$75.00

per 1M tokens

Cached Input

$1.50

per 1M tokens

Batch-Input

$7.50

per 1M tokens

Kontextfenster: 200K

Max. Output: 32,000 tokens

Wissensstand: 2025-03

VisionFunktionsaufrufFeinabstimmungJSON-ModusKostenlose Stufe

Vorteile

Best agentic coding model available
Superior deep analysis
Latest knowledge cutoff (2025-03)

Nachteile

Most expensive model in market ($15/$75)
No fine-tuning
Slower than Sonnet

Leistung

Ausgabegeschwindigkeit~45 tok/s

Rate-Limit2,000 RPM

Multimodal

BildeingabeBildausgabeAudioeingabeAudioausgabe

Benchmarks

SWE-bench Verified

72.7%

MMLU

89.2%

GPQA

76.8%

MATH

78.0%

Claude 4 Sonnet

Flagship

Coding, writing, analysis

Offizielle Preise

Wann verwenden: Best balance of quality and cost for writing, coding, and analysis production apps.

Upgrade-Highlights

◆SWE-bench: 63.8% — competitive with flagships at 1/5 the price
◆Batch API: $1.50/$7.50 — 50% savings for non-real-time workloads
◆200K context + 16K output — 2x output of Claude 3.5 Sonnet
◆Top-tier writing quality — best for content generation and editing
◆Price $3/$15 — replaced Claude 3.5 Sonnet as the default choice

Input-Preis

$3.00

per 1M tokens

Output-Preis

$15.00

per 1M tokens

Cached Input

$0.300

per 1M tokens

Batch-Input

$1.50

per 1M tokens

Kontextfenster: 200K

Max. Output: 16,384 tokens

Wissensstand: 2025-03

VisionFunktionsaufrufFeinabstimmungJSON-ModusKostenlose Stufe

Vorteile

Top-tier writing quality
5x cheaper than Opus
Excellent coding with 200K context

Nachteile

Lower max output than Opus
No fine-tuning
Slower than GPT-4o mini

Leistung

Ausgabegeschwindigkeit~70 tok/s

Rate-Limit4,000 RPM

Multimodal

BildeingabeBildausgabeAudioeingabeAudioausgabe

Benchmarks

SWE-bench Verified

63.8%

MMLU

88.0%

HumanEval

91.0%

MATH

72.5%

Claude 3.5 Haiku

Lite

Fast, affordable tasks

Offizielle Preise

Wann verwenden: For high-throughput text tasks where you want Anthropic quality on a budget.

Upgrade-Highlights

◆First budget Anthropic model with 200K context (was 100K in Claude 3 Haiku)
◆Prompt caching: $0.08/M — 90% savings for repeated system prompts
◆Batch API: $0.40/$2 — 50% savings for async processing
◆Claude's nuanced writing style at 1/4 the price of Sonnet
◆No vision — text-only; upgrade to Haiku 4.5 for multimodal

Input-Preis

$0.800

per 1M tokens

Output-Preis

$4.00

per 1M tokens

Cached Input

$0.080

per 1M tokens

Batch-Input

$0.400

per 1M tokens

Kontextfenster: 200K

Max. Output: 8,192 tokens

Wissensstand: 2024-07

VisionFunktionsaufrufFeinabstimmungJSON-ModusKostenlose Stufe

Vorteile

200K context at budget price
Claude's nuanced style at low cost
Good prompt caching savings

Nachteile

No vision support
No fine-tuning
8K max output is limiting

Leistung

Ausgabegeschwindigkeit~95 tok/s

Rate-Limit8,000 RPM

Multimodal

BildeingabeBildausgabeAudioeingabeAudioausgabe

Benchmarks

MMLU

83.0%

HumanEval

82.0%

MATH

63.5%

Nebeneinander-Vergleich

Modell	Stufe	Input	Output	Cached	Kontext	Max. Output
Claude Opus 4.8	Flagship	$5.00	$25.00	$0.500	1M	32,000
Claude Opus 4.7	Flagship	$5.00	$25.00	$0.500	1M	32,000
Claude Haiku 4.5	Lite	$1.00	$5.00	$0.100	200K	8,192
Claude 4 Opus	Flagship	$15.00	$75.00	$1.50	200K	32,000
Claude 4 Sonnet	Flagship	$3.00	$15.00	$0.300	200K	16,384
Claude 3.5 Haiku	Lite	$0.800	$4.00	$0.080	200K	8,192