Skip to content
Anthropic

Claude Haiku 4.5

budget

Anthropic•Released on 2025-10-01

Anthropic's fastest model in the Claude 4.5 family. Optimized for quick responses and high-throughput applications. Default fast model in Claude Code. Excellent for simple coding tasks, quick Q&A, and cost-sensitive batch processing.

79
Overall Score

Voice of the community

sample 634

“Haiku 4.5 in Claude Code is perfect for quick answers. I use it for simple refactors and questions, switch to Opus for complex work.”

Community2026-02-15

“The speed difference is noticeable. Haiku responds almost instantly while Opus can take several seconds for complex prompts.”

Community2026-01-20

“Don't use Haiku for anything requiring deep reasoning. It's fast but will miss important context in large codebases.”

Community2026-02-28

Core Specs

200K
Context Window
8K
Max Output
ReasoningOpen Sourcetextimage

Scenario Scores

Coding
#51 / 6780
Writing
#55 / 6774
Translation
#37 / 6782
Data Analysis
#57 / 6776
Conversation
#37 / 6782
Image Gen
—

Pros & Cons

Sentiment68% +24% ·8% −

Pros

  • +Fastest response in Claude family
  • +Affordable pricing ($1/$5 per MTok)
  • +200K context window
  • +Default fast option in Claude Code
  • +Great for high-volume batch processing

Cons

  • −Less capable than Sonnet/Opus for complex reasoning
  • −Smaller output limit (8K)
  • −Not ideal for multi-step agentic tasks
  • −May miss nuances in complex instructions

Pricing

Input (per 1M tokens)$1.00
Output (per 1M tokens)$5.00
Subscription$20/month
Free trial available
Updated on 2026-03-09

Get Started

1Visit the provider's website
2Create an account
3Start using the model

Benchmarks

mmlu86.5%
humanEval90.2%

Tools Supporting This Model

Claude Code
Anthropic
Claude Cowork
Anthropic
OpenClaw
OpenClaw

Reliability

SLA99.9%
Incidents (30d)8
Last Incident2026-06-10
与其他 Claude 模型共享基础设施；**5/18-5/19-5/20 Haiku 4.5 连续三日 elevated errors** (5/18 06:12-08:07 UTC 115min + 5/19 07:00-08:30 UTC 90min + 5/20 08:14-08:49 UTC 35min)，50h 内 3 次单模型 incident 构成日级周期模式 — 三起均集中在 06:00-09:00 UTC (亚太晚间 / 欧洲晨高峰 / 美东凌晨)，提示该时段触发因子；duration 趋势走低 (115min→90min→35min) 提示根因部分修复；按既定 SOP 单模型 moderate 仍不触发 Telegram (前例 5/18+5/19 同样未发)，下窗口若 5/21 同时段再现则升级 'reliability watch' Telegram；5/12 19:36-20:13 UTC ~37min Sonnet 4.6 + Haiku 4.5 双模型 elevated errors (moderate)；近期有稳定性波动记录 (4/28 51min + 4/29 60min Haiku 4.5 专属 elevated errors)；4/30 01:20-01:51 UTC 31min 全产品线不可用同样影响 Haiku 4.5 流量
View Status Page →

Compare with Others

Claude Haiku 4.5 vs Claude Fable 5 →Claude Haiku 4.5 vs Nex-N2-Pro →Claude Haiku 4.5 vs Nemotron 3 Ultra →