Claude Opus 5

Supported Functionality

Item	Specification
Input	Text, images
Output	Text
Context	1,000,000 tokens
Max Output	128,000 tokens; 300,000 tokens in Message Batches API beta
Vision	✓ Supported
Function Calling	✓ Supported

Description

Claude Opus 5 is an Anthropic language model released on July 24, 2026, with the official Claude API ID claude-opus-5. Built for complex agentic coding, enterprise knowledge work, and deep reasoning, it accepts text and images, produces text, and has a one-million-token context window with a standard 128,000-token maximum output. Anthropic recommends it as the starting point for evaluating most demanding workflows.

Compared with Opus 4.8, it improves long-horizon agents, complex reasoning, scientific analysis, computer use, and test-time compute scaling. Adaptive thinking is enabled by default, while effort controls reasoning depth; Anthropic lists more moderate comparative latency than Fable 5.1, making it suitable for regular high-capability work.

Key Capabilities

• Complex Agentic Coding: Builds across files, refactors, locates errors, and validates changes while coordinating tools and tests.
• Deep Reasoning: Decomposes ambiguous or constrained problems, compares options, and checks conclusions.
• Long-Horizon Execution: Tracks goals, state, and intermediate results across multistep enterprise processes.
• Long-Context Analysis: Connects code, documents, images, and past decisions within extensive input.
• Vision and Computer Use: Understands images and interfaces for browser and desktop tasks.
• Knowledge and Scientific Work: Helps organize research, analyze data, explore life-science questions, and support complex business decisions.
• Tools and Thinking Control: Uses tools for external actions and adjusts reasoning depth through adaptive thinking and effort.

Technical Strengths

Feature	Benefit
Adaptive Thinking by Default	Applies reasoning to difficult tasks while allowing it to be disabled under supported settings for simpler work.
Test-Time Compute Scaling	Higher effort can deepen analysis on difficult problems, enabling configuration through real-world evaluation.
One-Million-Token Context	Holds large codebases, long documents, and multistep project history together.
Long-Horizon Agent Capability	Combines planning, tools, and validation so fewer stages require manual handoff.
Joint Image and Text Input	Brings charts, screenshots, and document images into the same analysis.
Large Output and Batch Option	Supports 128K standard output and a higher beta limit under specific official batch conditions.

Frequently Asked Questions

How should I set effort for Claude Opus 5?

Anthropic recommends starting with the default high and then tuning against real task evaluations. Test higher settings for difficult reasoning and long agents, while checking whether lower settings already meet the acceptance criteria for routine work.

Is the 300K output limit available on every Opus 5 request?

No. Standard maximum output is 128K tokens; 300K is a beta capability of the official Message Batches API under its required beta setting, not the default for ordinary online requests.

How can Opus 5 avoid missing changes in a large repository?

Specify the target modules, test commands, and interfaces that must remain stable, then ask it to map dependencies and plan staged edits. Run tests and record validated files after each stage instead of requesting a whole-repository rewrite in one step.

Should I choose Opus 5 or Fable 5.1?

Anthropic recommends starting most evaluations with Opus 5. If it still misses requirements at higher effort on extended research, coding, or cross-app tasks, compare Fable 5.1 using identical inputs and acceptance criteria.

What should I test before using Opus 5 on LinkModel ?

Confirm the model ID, context, image input, tool calls, and request schema on LinkModel's live model page. Then test long tasks, output truncation, thinking parameters, and failure handling in the real workflow instead of copying Anthropic-only options unverified.