跳转到主要内容

Claude Haiku 5.5 (Max)

AnthropicClaudeProprietary

描述

Claude Haiku 5.5 is Anthropic's cheapest, fastest, and most capable small model to date, and the successor to Claude Haiku 4.5. It targets high-volume, cost-sensitive work (summaries, compaction, classification, database queries) and pairs with Opus 5.5 / Sonnet 5.5 as a coding subagent; it is especially strong for speed-sensitive tasks like live support and browser use. It is the first Haiku-class model with adjustable effort (low/medium/high/xhigh/max; Claude Platform default medium). It accepts text and image input (including file/PDF) with text output, supports multilingual and vision workloads, tool use, and adaptive thinking. Context window is 1M tokens with 128K max output; knowledge cutoff is June 2026. Claude Platform pricing is tiered by prompt length: prompts up to 100K tokens are $0.10/$0.50 per million input/output (cache read $0.01, 5m cache write $0.125); prompts over 100K are $0.50/$2.50 (cache read $0.05, 5m cache write $0.625). Catalog provider pricing stores the <=100K tier. Capability scores are self-reported from the Claude Haiku 5.5 System Card (Table 8.1.A and §8) and the launch announcement. Available on the Claude API as `claude-haiku-5-5`.

发布日期
2026-10-07
参数规模
—
上下文长度
1.0M
支持模态
image, pdf, text

能力雷达图

44
general
55
coding
50
reasoning
51
science
30
agents
60
multimodal

排行榜排名

领域#排名分数来源
智能体能力模型榜42
50.0
LS
代码能力榜15
93.0
AA
通用能力榜23
74.0
AA
数学推理39
86.0
LB
多模态榜83
50.0
LS
推理能力29
85.0
LB
科学能力49
76.0
AA

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

Agents

AA-Briefcase v1.11578.00 / 3000自报
Program Bench82.0%自报
OfficeQA73.5%自报
OfficeQA Pro60.3%自报

Biology

LatchBio SpatialBench Verified67.7%自报
Protocols Understanding (V2)64.1%自报
Protocols Troubleshooting58.0%自报
Medicinal Chemistry (ADME)58.0%自报
Protein Design — Library Ranking45.3%自报
Biomedical image analysis43.5%自报
Protein Design — Sequence Generation33.1%自报
Morphology-to-molecule matching25.4%自报

Code

BenchCAD (with Python tool)87.0%自报
BenchCAD67.0%自报
FrontierSWE V243.8%自报

General

GDPval-AA 2.11620.00 / 3000自报
Global-MMLU87.8%自报

Healthcare

HealthBench Professional (raw)71.0%自报
HealthBench Professional64.8%自报
HealthBench61.6%自报
PhysicianBench43.0%自报

Language

MILU87.6%自报

Multimodal

OSWorld 2.1 (partial)72.4%自报
OSWorld 2.1 (strict)37.1%自报

Reasoning

SWE-bench Multilingual83.7%自报
SWE-Bench ProPrinceton NLP (2024)64.8%自报
FrontierCode 1.1 (Extended)58.4%自报
Humanity's Last Exam (with tools)57.4%自报
FrontierCode 1.146.4%自报
Humanity's Last Exam (no tools, text-only)45.9%自报
Terminal-Bench 4.039.2%自报
SWE-Bench Multimodal30.7%自报
Terminal-Bench-Science 0.120.6%自报

Science

LatchBio SingleCellBench56.2%自报

Vision

Chartography86.2%自报
Chartography (no tools)46.4%自报

AA 评测指数

(Artificial Analysis)
Lcr(Artificial Analysis)
82.7
Scicode(UIUC + Argonne National Lab (2024))
55.0
Hle(Center for AI Safety + Scale AI (2025))
44.4
Intelligence Index(Artificial Analysis)
43.4
Terminalbench V4 0
32.8

LLM Stats 分类评分

(LLM Stats (zeroeval))
Productivity
100
Agents
100
Reasoning
100
General
100
Language
90
Multimodal
60
Science
60
Healthcare
60
Biology
60
Code
60
Vision
60
Math
50
Tool Calling
30

定价

输入价格$0.1 / 1M tokens
输出价格$0.5 / 1M tokens
混合价格(3:1)$0.2 / 1M tokens
缓存读取价格$0.01 / 1M tokens
缓存写入价格$0.125 / 1M tokens

速度

Tokens/秒227.9
首Token延迟292.43s
首回答延迟292.43s

供应商价格排行

供应商价格排行

1 个供应商

供应商输入输出
1Anthropic
$0
$0

比较该模型在不同 API 供应商之间的定价。

外部链接