# Claude Haiku 3.5 ![Organization](https://img.shields.io/badge/Org-Anthropic-d97757) ![Context](https://img.shields.io/badge/Context-200K%20tokens-1f6feb) ![Pricing](https://img.shields.io/badge/Price-%240.80%20%2F%20%244%20per%201M-2da44e) ![License](https://img.shields.io/badge/License-Proprietary-d29922) ![Modality](https://img.shields.io/badge/Modality-Text%20%2B%20Vision-8957e5) ![Released](https://img.shields.io/badge/Released-Oct%202024-555) > **Quick answer:** Claude Haiku 3.5 is Anthropic's fastest and most cost-efficient model in the Claude 3.x family, released October 22, 2024. It is designed for high-throughput, latency-sensitive production workloads — handling the same task types as Claude Sonnet and Opus but optimised for speed and cost at $0.80 input / $4.00 output per 1M tokens with a 200K-token context window. ## At a Glance **Where Claude Haiku 3.5 leads** - Lowest latency in the Claude 3.x family — fastest time-to-first-token - Most cost-efficient Claude 3.x model at $0.80/$4.00 per 1M tokens - 200K-token context window — same as Sonnet and Opus - Multimodal vision support - Prompt caching for further cost reduction on repeated contexts - Suitable for real-time applications: chatbots, classification, summarisation **Where it lags** - Lower capability ceiling than Claude Sonnet 3.5 and Sonnet 3.7 - SHADE-Arena score of 0.0 — note: this reflects well-aligned safety behaviour, not capability - No extended thinking mode - Superseded by Claude Haiku 4 for most new deployments **Best for:** Real-time customer interactions, high-volume batch classification, lightweight coding assistance, and any Claude-compatible pipeline where latency and cost are the primary constraints. ## What Claude Haiku 3.5 Is Claude Haiku 3.5 is Anthropic's small, fast tier of the Claude 3.5 release cycle, positioned for applications that need Claude-quality instruction following and safety at the lowest possible cost and latency. Like all Claude models, it is trained with Anthropic's Constitutional AI approach, which produces reliable instruction following and consistent refusal of harmful requests. The Haiku tier is designed for integration into user-facing products where response time matters: customer service bots, coding autocomplete, real-time document tagging, and API pipelines handling thousands of requests per minute. At $0.80/$4.00 per 1M tokens, it is one of the most affordable ways to access a multimodal, vision-capable language model from a major AI lab. Claude Haiku 3.5 retains the 200K-token context window of the full Claude 3.5 family, which distinguishes it from lower-end alternatives: it can process long documents and multi-session histories without chunking, even at the cost-efficient tier. Prompt caching further reduces costs for repeated system prompts or large reference documents. ## Specifications | Field | Value | |---|---| | Organization | Anthropic | | Parameters | Undisclosed | | Context window | 200,000 tokens | | Max output | 8,192 tokens | | Architecture | Transformer (Anthropic internal) | | License | Proprietary (API only) | | Release date | October 22, 2024 | | Knowledge cutoff | July 2024 (estimated) | | Modality | Text + Vision (multimodal) | ## Pricing | | Input (per 1M tokens) | Output (per 1M tokens) | |---|---|---| | Anthropic API | $0.80 | $4.00 | Prompt caching available: cache writes at 1.25× the input rate, cache reads at 90% discount. Pricing per [Anthropic pricing page](https://www.anthropic.com/pricing). ## Context Window Claude Haiku 3.5 has a **200,000-token context window** — roughly **150 pages** of text in a single request. This is the same as Claude Sonnet 3.7, making Haiku 3.5 suitable for long-document tasks even at the cost-efficient tier. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [SHADE-Arena](/benchmarks/shade-arena) | 0.0 overall success | [Anthropic research post](https://www.anthropic.com/research/shade-arena-sabotage-monitoring) | 2025-06 | *Note: The SHADE-Arena score of 0.0 indicates Claude Haiku 3.5 did not engage in sabotage behaviour — this is expected and reflects well-aligned safety training, not limited capability.* ## Claude Haiku 3.5 vs Alternatives | Model | Context | Speed tier | Price (in/out per 1M) | |---|---|---|---| | **Claude Haiku 3.5** | 200K | Fastest (Claude 3.x) | $0.80 / $4 | | [Claude Sonnet 3.7](/models/anthropic/claude-3-7-sonnet) | 200K | Balanced | $3 / $15 | | [Gemini 2.5 Flash](/models/google/gemini-2-5-flash) | 1M | Fast | $0.15 / $0.60 | | [o4-mini](/models/openai/o4-mini) | 200K | Reasoning | $1.10 / $4.40 | Claude Haiku 3.5's main competitor is Gemini 2.5 Flash, which is cheaper at $0.15/$0.60 and offers a 5× larger context window. Haiku 3.5's advantages: lower latency, Anthropic's constitutional AI safety profile, and seamless compatibility with Claude API integrations already in production. ## How Claude Haiku 3.5 Performs on Real Agent Tasks Claude Haiku 3.5's SHADE-Arena score of 0.0 (the model does not engage in sabotage behaviour) is the expected result for a well-aligned model. It is not a measure of task capability. For agent builders, Haiku 3.5 is the appropriate choice for the "fast lane" of a tiered agent architecture — handling routine classification, data extraction, and summarisation tasks while routing complex or ambiguous cases to Claude Sonnet or Opus. This approach maximises throughput while keeping overall pipeline cost low. ## Use Claude Haiku 3.5 via API ```python import anthropic client = anthropic.Anthropic(api_key="YOUR_API_KEY") response = client.messages.create( model="claude-haiku-3-5-20241022", max_tokens=1024, messages=[{"role": "user", "content": "Classify this customer support ticket as: billing, technical, or account..."}], ) print(response.content[0].text) ``` ## Related Models & Benchmarks - Compare: [Claude 3.7 Sonnet](/models/anthropic/claude-3-7-sonnet) · [Claude Haiku 4.5](/models/anthropic/claude-haiku-4-5) · [Gemini 2.5 Flash](/models/google/gemini-2-5-flash) - Benchmarks: [SHADE-Arena](/benchmarks/shade-arena) ## Frequently Asked Questions

What is Claude Haiku 3.5?

Claude Haiku 3.5 is Anthropic's fastest and most cost-efficient model in the Claude 3.x family, released October 22, 2024. It is designed for high-throughput, latency-sensitive applications at $0.80/$4.00 per 1M tokens with a 200K-token context window.

What is Claude Haiku 3.5's context window?

Claude Haiku 3.5 supports a 200,000-token context window — the same as Claude Sonnet 3.7 — with up to 8,192 output tokens.

How much does Claude Haiku 3.5 cost?

Claude Haiku 3.5 costs $0.80 per 1M input tokens and $4.00 per 1M output tokens. Prompt caching provides a 90% discount on cache reads, reducing costs further for repeated contexts.

Is Claude Haiku 3.5 open source?

No. Claude Haiku 3.5 is a proprietary, API-only model. There are no open weights.

What is the difference between Claude Haiku 3.5 and Claude Sonnet 3.7?

Claude Haiku 3.5 ($0.80/$4 per 1M) is optimised for speed and cost, with lower capability ceiling. Claude Sonnet 3.7 ($3/$15 per 1M) includes extended thinking mode and is significantly stronger on complex reasoning and coding tasks. Haiku 3.5 is ~4× cheaper but ~4× less capable on frontier tasks.

What is Claude Haiku 3.5's knowledge cutoff?

Claude Haiku 3.5's training knowledge cutoff is estimated at July 2024.

--- *Specs and scores sourced from Anthropic's official Claude Haiku 3.5 announcement (October 22, 2024) and Benchgen evaluations; third-party benchmark scores attributed inline. Pricing cited to the Anthropic pricing page. Last updated 2026-07-23.*

Claude Haiku 3.5 — Evaluations

Evaluations