Robots Atlas>ROBOTS ATLAS
Claude Opus 4.8
AI Models›Claude

Claude Opus 4.8

4.8 · Family: Claude
Anthropic’s latest flagship model in the Claude Opus family, optimized for agentic tasks, coding, and long multi-step sessions.
✓ Active✓ Public accessLLMMultimodalReasoning modelTool-using model📁 Claude
Context window
1M
tokens
Release date
28 May 2026
🏢AnthropicProducer
Access:APIHostedDeployment:☁ Cloud

Overview

Claude Opus 4.8 is Anthropic’s flagship Claude-family model released on May 28, 2026, succeeding Claude Opus 4.7. Standard pricing matches the predecessor ($5/$25 per million input/output tokens), while fast mode is three times cheaper than on Opus 4.7 and 4.6 ($10/$50 per million tokens) at 2.5× the generation speed.

Anthropic reports that Opus 4.8 is roughly four times less likely than its predecessor to let flaws in its own code pass unremarked, and the alignment assessment shows substantially lower rates of misaligned behavior (e.g., deception) than Opus 4.7. The release ships alongside dynamic workflows in Claude Code (hundreds of parallel subagents) and an effort control (low/medium/high/extra/max) in claude.ai and Cowork.

The model is available via the claude-opus-4-8 identifier in the Claude API, in claude.ai apps (Free, Pro, Max, Team, Enterprise plans), and on partner cloud platforms.

Classification
LLMMultimodalReasoning modelTool-using model
Family: Claude
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
📏 Context: 1M
✓ Tools
📥 Input: text, image, documents
Platforms
Anthropic Claude APIAmazon BedrockVertex AIMicrosoft Azure AI Foundry

Technical specification

Context window
1M
tokens
Features:✓ Tool use
Modalities
⬇ Input
textimagedocuments
⬆ Output
textcodestructured_data

Capabilities and applications

Native model capabilities
Reasoning
The model's ability to reason logically and solve complex problems.
Category: reasoning
Multi-step reasoning
Carrying out multi-step chains of reasoning across long, complex tasks.
Category: reasoning
Long context
Maintaining coherence and focus across very long input context.
Category: language
Coding
Generating, analysing and modifying source code.
Category: coding
Structured output
Producing data in structured formats such as JSON.
Category: structured_generation
Image understanding
Analysing and interpreting the content of images.
Category: vision
Chart understanding
Reading and interpreting charts, tables and diagrams.
Category: vision
OCR
Recognising text within images and documents.
Category: vision
Multilingual
Understanding and generating text in many languages.
Category: language
Planning
Forming and executing action plans for complex tasks.
Category: planning

Benchmark results

8 benchmarks
SWE-Bench Pro
agentic coding
69.2%%
📄 Anthropic, Introducing Claude Opus 4.8 (2026-05-28)
Terminal-Bench 2.1
Terminus-2 harness
74.6%%
📄 Anthropic, Introducing Claude Opus 4.8 (2026-05-28)
Humanity’s Last Exam (no tools)
49.8%%
📄 Anthropic, Introducing Claude Opus 4.8 (2026-05-28)
Humanity’s Last Exam (with tools)
57.9%%
📄 Anthropic, Introducing Claude Opus 4.8 (2026-05-28)
OSWorld-Verified
agentic computer use
83.4%%
📄 Anthropic, Introducing Claude Opus 4.8 (2026-05-28)
GDPval-AA
knowledge work
1890
📄 Anthropic, Introducing Claude Opus 4.8 (2026-05-28)
Finance Agent v2
53.9%%
📄 Anthropic, Introducing Claude Opus 4.8 (2026-05-28)
Online-Mind2Web
84%%
📄 Browserbase (Miguel Gonzalez quote, Anthropic blog 2026-05-28)

Pricing

Technical architecture

Deployment and security

🔒 Security / Enterprise
✓ Verified enterprise information

The model underwent a full set of evaluations under Anthropic's Responsible Scaling Policy (RSP), including assessments of cyber, chemical, and biological risk, AI R&D autonomy, and misalignment. Anthropic's conclusions: catastrophic risk remains low with current mitigations. Opus 4.8 demonstrates a significantly improved alignment profile relative to Opus 4.7 (including ~4× lower rate of failing to report flaws in its own code).

Anthropic conducted its first week-long live bug bounty for prompt injection. In agentic contexts, Opus 4.8 is slightly less resistant to prompt injection than Opus 4.7, but Anthropic reports that its own safeguards close this gap in production.
Updated: 28 May 2026↗ Security documentation

Sources and related pages