Skip to main content
Skip to main content

Claude Haiku 4.5

Claude 4.5 Haiku is Anthropic's fast and cost-efficient member of the Claude 4.5 family, released October 15, 2025. It is the model you reach for when Sonnet-class quality is overkill and per-token cost is the deciding factor.

  • 99.9% SLA
  • Official Discounts
  • Pay-as-you-go
  • Fast & Low Latency
Brand:AnthropicInput token: $1.00/MOutput token: $5.00/MInput cache read: $0.10/MInput cache write: $1.25/M

Claude 4.5 Haiku: Near-Frontier Speed at a Fifth of the Price

Claude 4.5 Haiku (catalog slug anthropic-claude-4-5-haiku-20251001) is Anthropic's fast and cost-efficient member of the Claude 4.5 family, released October 15, 2025. It is the model you reach for when Sonnet-class quality is overkill and per-token cost is the deciding factor.

Here's the short version: 200K context, 64K output, $1/$5 per 1M tokens, and 73.3% on SWE-bench Verified - Anthropic positions it at roughly 90% of Sonnet 4.5's quality for about 20% of the price. It is text-native, fast, and genuinely good at coding for its tier. The honest caveats: it is not a frontier reasoning model, and "90% of Sonnet" is a vendor positioning claim - test on your own workload before believing it.

This guide covers model overview, core features, technical specifications, capability comparison, core advantages, recommended use cases, example prompts, selection recommendations, and my verdict.

Quick Facts

AttributeValue
Catalog sluganthropic-claude-4-5-haiku-20251001
Model IDclaude-haiku-4-5-20251001
DeveloperAnthropic
ReleaseOctober 15, 2025
Context / output200K / 64K tokens
Pricing$1.00 / $5.00 per 1M tokens
SWE-bench Verified73.3%
Positioning~90% of Sonnet 4.5 at ~20% of cost
ModalityText

Table of Contents

  1. Model Overview
  2. Core Features
  3. Technical Specifications
  4. Capability Comparison
  5. Core Advantages
  6. Recommended Use Cases
  7. Example Prompts
  8. Selection Recommendations
  9. The Bottom Line
  10. FAQ
  11. Sources & Further Reading

1. Model Overview

Anthropic launched Claude Haiku 4.5 on October 15, 2025 as the speed tier of the Claude 4.5 generation. The family's Haiku role has always been volume: low latency, low price, good-enough quality. What changed with 4.5 is how much quality "good enough" means - 73.3% on SWE-bench Verified made it one of the best coding models in the world at launch, at a price point that changed how teams budget for agentic workloads.

The release shipped alongside a system card documenting safety evaluations, and it landed on the Anthropic API, Claude Code, and cloud marketplaces such as AWS Bedrock and Google Vertex AI. In the catalog, the model is also available under a legacy snapshot ID (...-legacy-759f9601) for integrations that need pinned behavior.

2. Core Features

Fast execution. The Haiku tier is Anthropic's lowest-latency family.

Aggressive pricing. $1 input / $5 output per 1M tokens - roughly a fifth of Sonnet 4.5.

Coding strength. 73.3% SWE-bench Verified, competitive with far pricier models.

200K context. Long files, repos, and document batches fit in one call.

64K output. Room for large generated artifacts.

Text-native. The 4.5 line is text-focused; pair it with a vision model when images are required.

Ecosystem coverage. Anthropic API, Claude Code, Bedrock, Vertex AI, and major router platforms.

3. Technical Specifications

SpecificationDetail
Model IDclaude-haiku-4-5-20251001
Context window200K tokens
Max output64K tokens
Input price$1.00 per 1M tokens
Output price$5.00 per 1M tokens
SWE-bench Verified73.3%
Release dateOctober 15, 2025
ModalityText (no vision input)
AvailabilityAnthropic API, Claude Code, Bedrock, Vertex AI

Note: benchmark figures are Anthropic-reported at release. Independent leaderboards (e.g., Artificial Analysis) track the model's live performance and pricing across providers.

4. Capability Comparison

Benchmark / attributeHaiku 4.5Sonnet 4.5Opus 4.5Opus 4.6
SWE-bench Verified73.3%77.2%80.9%higher tier
Context window200K200K200K1M
Input price / 1M$1.00$3.00$5.00$5.00
Output price / 1M$5.00$15.00$25.00$25.00
Best roleVolume / speedBalancedFrontierFrontier + long context

Reading the table honestly: Haiku 4.5 gives you the best cost-per-task in the family. Opus-class models remain measurably better on hard reasoning; Haiku's job is to make that gap affordable at scale.

5. Core Advantages

  1. Best cost-per-task in the Claude family. $1/$5 makes agentic volume viable.
  2. Speed. Low latency for interactive and streaming use.
  3. Strong coding for the tier. 73.3% SWE-bench Verified at release.
  4. Long context. 200K covers serious code and document workloads.
  5. Mature tooling. Claude Code, Bedrock, and Vertex integration.
  6. Predictable scaling. Same family behavior at a fraction of Opus cost.

6. Recommended Use Cases

  • High-volume chat and support: classification, extraction, summarization.
  • Coding assistants: autocomplete, review triage, test generation.
  • Batch document processing: 200K context for long inputs.
  • Agent workflows on a budget: tool-calling pipelines where tokens dominate cost.
  • Drafting and rewriting: marketing, email, and content at scale.

7. Example Prompts

1. Extraction at scale

text
Extract all entities (person, org, date, amount) from this contract batch into
JSON. Preserve exact strings; do not paraphrase.

2. Code review triage

text
Review this diff. Output a table: file, issue type (bug/security/perf/style),
severity, and a concrete fix. Ignore formatting nits.

3. Agentic task with tools

text
You have tools: search_docs, get_issue, create_pr. Triage this bug report:
verify it exists, find the likely root cause in the docs, and draft a PR
description. Keep token use minimal.

4. Long-document summary

text
Summarize this 100-page spec into a two-page handoff: goals, architecture,
constraints, open questions, and owners.

8. Selection Recommendations

Choose Claude 4.5 Haiku if:

  • Cost per task dominates your model decision.
  • Your workload is high-volume, latency-sensitive, or both.
  • You need coding competence but not frontier reasoning.
  • You want one text model for extraction, classification, and drafting.

Choose Sonnet 4.5 if:

  • You need a step up in reasoning and are comfortable at 3x the token cost.
  • Quality differences show up in your A/B tests on real workloads.

Choose Opus 4.5 / 4.6 if:

  • The task is hard reasoning, long-horizon agents, or 1M-context (4.6).
  • Token price is secondary to getting the answer right.

9. The Bottom Line

Verdict: The sensible default for high-volume text work in 2026. Claude 4.5 Haiku delivers near-frontier coding strength at a fifth of the price of its bigger siblings, with 200K context and fast response times. It is not the model for the hardest reasoning problems - that is what Opus is for - but for cost-conscious production workloads, it is arguably the most rational choice in Anthropic's lineup.

Sources & Further Reading

Benchmark figures are vendor-reported at release unless noted. Independent results can differ; validate against your own workload before committing.