What is the pricing for Claude Haiku 5.5?

Standard pricing is $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Prompts exceeding 100,000 tokens cost $0.50 per million input tokens and $2.50 per million output tokens.

How does Haiku 5.5 perform compared to Haiku 4.5?

Haiku 5.5 shows significant improvements, scoring 72.4% on the OSWorld 2.1 offline subset compared to 15.7% for Haiku 4.5, and achieving higher scores on GDPval-AA and AA-Briefcase benchmarks.

What are the main constraints of Claude Haiku 5.5?

Haiku 5.5 includes stricter safety guardrails regarding cybersecurity penetration testing and biological data, making it more limited in those specific research areas compared to previous versions.

What is the context window for Claude Haiku 5.5?

Claude Haiku 5.5 features a 1,000,000 token context window with a maximum output limit of 128,000 tokens.

Claude Haiku 5.5 Features Pricing and Benchmarks

Claude Haiku 5.5 Features and Pricing. A macro close-up of a single, vibrant sapphire-colored fiber optic cable snaking through a dark, brushed-aluminum server rack, illuminated by the cold, rhythmic pulse of an emerald status light — GPUniq

Data as of October 8, 20263 sources

Drafted with AI tools and edited by a human. The figures and conclusions were checked by a GPUniq editor.

TL;DR

Claude Haiku 5.5 scores 72.4% on the OSWorld 2.1 offline subset compared to 15.7% for Haiku 4.5. Standard input pricing starts at $0.10 per million tokens for prompts up to 100,000 tokens. Inputs exceeding 100,000 tokens cost 5x more at $0.50 per million tokens.

How Fast Is Haiku 5.5?

Claude Haiku 5.5 is optimized for classification, extraction, routing, and summarization. It includes an adaptive thinking parameter controlled via an "effort" setting.

Spec	Capacity
Context Window	1,000,000 tokens
Maximum Output	128,000 tokens
Input Modalities	Text, Image
Output Modality	Text

The 128,000 token output limit allows long sequence generation without hitting context caps mid-task. The model has a knowledge cutoff of June 2026.

How Much Does Haiku 5.5 Cost?

Anthropic uses two pricing tiers based on total prompt length. Prompts up to 100,000 tokens cost $0.10 per million input tokens and $0.50 per million output tokens.

Prompts exceeding 100,000 tokens trigger long-context pricing. Input rates increase to $0.50 per million tokens. Output rates increase to $2.50 per million tokens.

Tier	Input Rate per M Tokens	Output Rate per M Tokens
Standard (<=100K tokens)	$0.10	$0.50
Long Context (>100K tokens)	$0.50	$2.50
Cache Write (5 min cache)	$0.125	N/A
Cache Read	$0.010	N/A

Writing to the 5-minute cache costs $0.125 per million tokens. Reading cached tokens costs $0.010 per million tokens. Keeping prompts under 100,000 tokens avoids the 5x price increase on baseline API calls.

How Does It Benchmark Against 4.5?

Haiku 5.5 shows performance improvements over Haiku 4.5 across knowledge work and computer use benchmarks.

Evaluation Benchmark	Haiku 4.5 Score	Haiku 5.5 Score
GDPval-AA v2.1	735	1620
AA-Briefcase v1.1	614	1578
OSWorld 2.1 (Offline subset)	15.7%	72.4%

On GDPval-AA v2.1, Haiku 5.5 reaches 1620 points against 735 for Haiku 4.5. On AA-Briefcase v1.1, Haiku 5.5 reaches 1578 points compared to 614 for Haiku 4.5. Anthropic notes that Haiku 5.5 remains less suited for very large agentic coding tasks than Sonnet or Opus models.

Can Haiku 5.5 Handle Computer Use?

Haiku 5.5 supports text and image inputs to generate text outputs. On the OSWorld 2.1 offline subset evaluation for computer use, Haiku 5.5 scores 72.4%, up from 15.7% for Haiku 4.5.

OSWorld 2.1 Offline Subset Score:
Haiku 4.5 | [15.7%]
Haiku 5.5 | [72.4%]

The model handles visual inputs alongside text prompts for automated workflows. It is available via the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS.

What Are the Primary Guardrails?

Anthropic added specific safety constraints to Haiku 5.5. These safeguards target cybersecurity tasks and biological data risks:

Cybersecurity safeguards block commands for penetration testing.
Safeguards restrict responses to biology-related queries.
Restrictions make the model more limited than Haiku 4.5 for security auditing workflows.

These controls limit automated misuse in sensitive domains while restricting legitimate security research capabilities.

Methodology

Facts are compiled from the vendor's own announcement and documentation, linked below, at the time of writing; where the vendor hasn't published a number it is marked as such. GPUniq prices, when the model is already available here, come from our live catalog.