AI Agent Hub
Back to models
Anthropic: Claude Opus 5 logo

Anthropic: Claude Opus 5

Closed Source anthropic Released 2026-07-24
-- 1M context Proprietary

About this model

Claude Opus 5 is Anthropic's July 2026 flagship model for agentic coding, computer use, and long-horizon knowledge work. It is positioned as a substantial upgrade over Claude Opus 4.8, delivering near-frontier intelligence relative to Claude Fable 5 at roughly half the API cost ($5 per million input tokens and $25 per million output tokens). The model ships with adaptive extended thinking enabled by default, configurable effort from low through max, and is available on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry under the identifier claude-opus-5.

Anthropic's system card highlights state-of-the-art or leading results on software engineering and agent benchmarks, including SWE-bench Verified (96.0%), SWE-bench Pro (79.2%), FrontierBench v0.1 (43.3%), BrowseComp (90.8%), and GDPval-AA v2 (Elo 1861 at max effort). Opus 5 also improves computer-use scores on OSWorld 2.0, business workflow completion on AutomationBench, and scientific reasoning on GPQA Diamond and Humanity's Last Exam, while remaining behind Claude Mythos 5 on the highest-risk offensive cybersecurity and biology evaluations.

The model supports a 1 million-token context window (default and maximum), up to 128K tokens of output per request, and multimodal input of text and images with text-only responses. Knowledge and training data cut off in May 2026. It is designed for production agents, multi-step tool use, terminal and IDE workflows, deep research, and professional document tasks rather than on-premise self-hosting.

Benchmark Scores

ARC
97.5
HLE
63.6
MMLU
92.5
DeepSWE
73.65
MATH-500
97.6
MMLU-Pro
91.6
MMMU-Pro
84.7
AIME-2025
100.0
GDPval-AA
1861.0
LiveBench
80.08
BrowseComp
90.8
SimpleBench
80.6
GPQA-Diamond
93.7
HLE-Verified
54.4
Context-Arena
97.72
LiveCodeBench
89.0
NL2Repo-Bench
75.3
SWE-Bench-Pro
79.2
AutomationBench
50.3
Agents-Last-Exam
32.2
Creative-Writing
2132.6
SWE-Bench-Verified
96.0
Terminal-Bench-2.1
89.1
Terminal-Bench-3.0
43.3
Toolathlon-Verified
80.6
SWE-Bench-Multilingual
89.5

Technical Specs

  • Architecture: Transformer
  • Context Window: 1,000,000 tokens
  • Input Modalities: text, image

Hardware Requirements

  • Compute: API only

Pricing

Input Output Currency
5.00 / 1M tokens 25.00 / 1M tokens USD