跳转到主要内容

Claude Sonnet 5 (Max)

AnthropicClaudeProprietary

描述

Claude Sonnet 5 is Anthropic's most agentic Sonnet-class model, an upgrade to Sonnet 4.6 that narrows the gap to Opus 4.8 on reasoning, tool use, coding, computer use, and knowledge work while staying lower priced. It plans, uses tools like browsers and terminals, and runs autonomously for long-horizon tasks. Capability gains include SWE-Bench Verified (85.2%), SWE-Bench Pro (63.2%), SWE-Bench Multilingual (78.3%), Terminal-Bench 2.1 (80.4%), OSWorld-Verified (81.2%), BrowseComp (84.7% single-agent, 86.6% multi-agent), Humanity's Last Exam with tools (57.4%), USAMO 2026 (79.5%), GDPval-AA v2 (1618 Elo), HealthBench Professional (57.8%), and FrontierCode v1 (38.8%). It supports adaptive thinking with selectable effort levels up to 'extra high' (xhigh) and a 1M-token context window with context compaction. The safety assessment found lower rates of misaligned behavior, hallucination, and sycophancy than Sonnet 4.6, with improved prompt-injection robustness; it ships with cyber safeguards enabled by default and uses an updated tokenizer (input maps to roughly 1.0-1.35x more tokens than Sonnet 4.6). Default model on Free and Pro plans and available to Max, Team, and Enterprise users, in Claude Code, and on the Claude Platform. Launches with introductory pricing of $2/$10 per million input/output tokens through August 31, 2026, then $3/$15. Available via the Claude API as `claude-sonnet-5`.

发布日期
2026-06-30
参数规模
—
上下文长度
1.0M
支持模态
image, pdf, text

能力雷达图

2
general
50
coding
12
reasoning
51
science估算
40
agents
80
multimodal

缺少专门科学评测时,Science 由 LLM Stats 科学得分或推理能力估算。

排行榜排名

领域#排名分数来源
智能体能力模型榜122
37.0
LS
代码能力榜36
75.0
LB
多模态榜54
56.0
LS
推理能力19
89.0
LB

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

Agents

OfficeQA Pro59.4%自报
Toolathlon54.3%自报
AutomationBench13.5%自报

Code

DeepSWE 1.154.0%
BenchCAD37.3%自报

General

GDP.pdf81.6%自报

Healthcare

HealthBench Professional57.8%自报
LifeSci Diamond (2026-11)1.8%

Legal

Legal Agent Benchmark5.8%自报

Math

USAMO 202633.39 / 42自报
ArXivMath72.2%自报

Multimodal

OSWorld-Verified81.2%自报

Reasoning

CharXiv-R88.3%自报
ChartMuseum86.7%自报
SWE-Bench Verified85.2%自报
BrowseCompOpenAI (2025)84.7%自报
Terminal-Bench 2.0Stanford × Laude Institute (2026)80.4%自报
SWE-bench Multilingual78.3%自报
SWE-Bench ProPrinceton NLP (2024)63.2%自报
Humanity's Last Exam57.4%自报
FrontierCode 1.142.7%
FrontierCode38.8%自报
SWE-Bench Multimodal28.1%自报
Terminal-Bench 4.012.4%

AA 评测指数

(Artificial Analysis)

暂无 AA 评测数据

LLM Stats 分类评分

(LLM Stats (zeroeval))
Math
12
Reasoning
2
General
2
Frontend Development
90
Search
80
Multimodal
60
Healthcare
60
Vision
60
Agents
50
Code
50
Tool Calling
40
Legal
10

定价

输入价格$2 / 1M tokens
输出价格$10 / 1M tokens
混合价格(3:1)$4 / 1M tokens
缓存读取价格$0.2 / 1M tokens
缓存写入价格$2.5 / 1M tokens

速度

Tokens/秒79.0
首Token延迟167.06s
首回答延迟167.06s

供应商价格排行

供应商价格排行

2 个供应商

最便宜: Anthropic最贵: DeepInfra
供应商输入输出
1Anthropic主要
$2
$10
2DeepInfra
$3
$15

比较该模型在不同 API 供应商之间的定价。

外部链接