News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
imgClaude 3 Haiku

Claude 3 Haiku

Anthropic's fastest and lowest-cost model in the Claude 3 family

Claude 3 Haiku is a language model from Anthropic, released in March 2024 as the smallest and fastest member of the Claude 3 family. The company positions it for high-volume enterprise work such as document analysis and customer support, with vision input and low pricing.

Share on WhatsApp

  • Fastest and cheapest model in the Claude 3 family, per Anthropic
  • Processes about 21K tokens (roughly 30 pages) per second under 32K-token prompts
  • Anthropic claims 3x faster than peers for most workloads
  • Handles text and images; 2,500 images for one US dollar at the company's estimate
  • Available via Claude API, claude.ai, Amazon Bedrock and other cloud platforms

What it is

Claude 3 Haiku is a general-purpose language model with image understanding, made by Anthropic. It arrived in March 2024 alongside the larger Sonnet and Opus models in the same family. Anthropic describes it as the quickest and cheapest option in its performance class, aimed at work that involves large volumes of material: reading filings, contracts and legal cases, answering customer queries, and running many small jobs at once. It handles both text and images and returns text. The company reports throughput of 21,000 tokens, roughly 30 pages, per second for prompts below 32,000 tokens.

How it came about

Anthropic announced Haiku as the third member of its Claude 3 line, joining Sonnet and Opus. The launch post stresses speed and price rather than raw capability, targeting business users who process many documents or need quick chat replies. At release the model went live in the Claude API and for Claude Pro subscribers on claude.ai, and on Amazon Bedrock, with Google Cloud Vertex AI listed as coming soon. A later Anthropic notice says Claude is now available in Microsoft Foundry and Microsoft 365 Copilot.

Inside the model

Anthropic publishes little technical detail about Haiku's internals in this announcement: no parameter count, layer count, training corpus or compute figures are given. The company does describe its serving behaviour. Prompts under 32,000 tokens are ingested at about 21,000 tokens per second; longer prompts may run 30 to 60 per cent slower, a gap the company said it expected to narrow. Images add latency. Pricing uses a 1:5 ratio between input and output tokens, reflecting workloads with long prompts. Safety work includes testing against harmful outputs and jailbreaks, plus monitoring, endpoint hardening, encryption, access controls, audits and penetration testing.

What it can do

Anthropic claims strong results on industry benchmarks and state-of-the-art vision, without listing scores. The emphasis is on throughput and cost: the company says Haiku is three times faster than comparable models for most workloads and costs half as much as others in its tier. Illustrative figures given are 400 US Supreme Court cases, each estimated at 10,000 tokens, or 2,500 images at about 1,600 tokens each, processed for one US dollar. Image handling and very long prompts can slow it down.

How to use it

Haiku is offered through the Claude API and to Claude Pro subscribers on claude.ai. It is also on Amazon Bedrock, with Google Cloud Vertex AI announced as forthcoming, and Claude has since become available in Microsoft Foundry and Microsoft 365 Copilot. The weights are not released; access is through Anthropic's services and cloud platforms.

Safety and testing

Anthropic says it tests models to reduce harmful outputs and jailbreaks, and adds system monitoring, endpoint hardening, secure coding, data encryption and strict access controls. It runs regular security audits and works with penetration testers to find vulnerabilities. Further detail is directed to the Claude 3 model card.

Questions people ask about Claude 3 Haiku

What is Claude 3 Haiku for?

Anthropic built it for high-volume, latency-sensitive work: scanning large document sets, customer support chat, and running many small tasks together. It is the speed-and-price member of the Claude 3 family rather than the most capable one.

How fast is it?

Anthropic reports 21,000 tokens, about 30 pages, ingested per second for prompts under 32,000 tokens, and says it is three times faster than comparable models for most workloads. Longer prompts can be 30 to 60 per cent slower, and images add latency.

Can it read images?

Yes. Haiku accepts text and images and returns text. Anthropic estimates each image at around 1,600 tokens and says 2,500 images can be processed for one US dollar.

How do I get access to it?

Through the Claude API, a Claude Pro subscription on claude.ai, and Amazon Bedrock. Anthropic said Google Cloud Vertex AI support was coming, and Claude is now available in Microsoft Foundry and Microsoft 365 Copilot. The model weights are not openly released.

How large is the model?

Anthropic has not published a parameter count for Haiku in this announcement, nor details of its training data or compute. What the company does give is throughput, pricing structure and safety measures.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Claude 3 Haiku yet, so the facts on this page come from the sources listed above: anthropic.com/news/claude-3-haiku. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.