跳轉到主要內容

Claude Mythos Preview

AnthropicClaudeProprietary

描述

Claude Mythos Preview is an unreleased general-purpose frontier model from Anthropic, a new tier above Opus (internal codename 'Capybara'). It identified thousands of zero-day vulnerabilities across every major operating system and web browser as part of Project Glasswing, a cross-industry cybersecurity initiative with 12 partners including AWS, Apple, Microsoft, and Google. State-of-the-art on SWE-bench Verified (93.9%), GPQA Diamond (94.6%), USAMO (97.6%), Terminal-Bench 2.0 (82.0%), CyberGym (83.1%), and Cybench (100% pass@1, saturated). Represents a 4.3x increase over the previous trendline for model performance. Deployed under ASL-3 Standard. Best-aligned Claude model to date per Anthropic's risk report, with the first-ever 24-hour internal alignment review before deployment. Not planned for general availability. Pricing for participants: $25/$125 per million tokens (input/output). 244-page system card.

發布日期
參數規模
上下文長度
支援模態
image, text

能力雷達圖

90
general
80
coding
80
reasoning
77
science估算
80
agents
90
multimodal

Science 在缺少專門科學評測時使用推理能力代理估算。

排行榜排名

領域#排名分數來源
智能体与工具3
79.0
LS
多模态榜3
93.0
LS

基準測試分數 (LLM Stats)

Agents

CyBench100.0%自報
BrowseComp86.9%自報
CyberGym83.1%自報
Terminal-Bench 2.082.0%自報
OSWorld-Verified79.6%自報
SWE-Bench Pro77.8%自報
SWE-Bench Multimodal59.0%自報

Biology

GPQA94.6%自報

Code

SWE-Bench Verified93.9%自報
SWE-bench Multilingual87.3%自報

General

MMMLU92.7%自報

Healthcare

FigQA89.0%自報

Long Context

Graphwalks BFS >128k80.0%自報

Math

USAMO2597.6%自報
Humanity's Last Exam64.7%自報

Multimodal

CharXiv-R93.2%自報

AA 評測指數

暫無 AA 評測資料

LLM Stats 分類評分

Biology
90
Chemistry
90
Frontend Development
90
General
90
Healthcare
90
Language
90
Multimodal
90
Physics
90
Reasoning
90
Safety
90
Search
90
Spatial Reasoning
80
Tool Calling
80
Vision
80
Agents
80
Code
80
Long Context
80
Math
80

定價

暫無定價資料

速度

暫無速度資料

可用提供商

(LS 內部計價單位)

暫無提供商資料

外部連結