The History of AI

Claude Opus 4.5

Era 5 · Generative AI, reasoning, and agents (2022–2026) › Frontier models

Lineage
Neurons and networks
Kind
Implementation

Who

Anthropic

What happened

Anthropic releases Claude Opus 4.5, which it presents as the best model in the world for coding, agents, and computer use. It costs $5 per million input tokens and $25 per million output tokens.

Why it mattered

A high-end model built to work as an agent on long tasks, with a setting that controls how much effort it puts into each one.

The full story

By late 2025 the labs were competing above all on coding and on agents, systems that chain actions together to complete a task. On November 24, six days after Gemini 3 Pro, Anthropic released Claude Opus 4.5 and presented it as the best model in the world for coding, agents, and computer use.

Anthropic's figures stressed efficiency. At medium effort, Opus 4.5 matched Claude Sonnet 4.5's best score on SWE-bench Verified, a benchmark of real programming issues, while using 76 percent fewer output tokens; at maximum effort, it beat that score by 4.3 points. "Effort" is a setting that controls how much work the model puts into each task.

For long-running agents, Anthropic added tools for compacting context and for coordinating teams of subagents. According to the company, it was also its best-aligned model and the hardest to trick with instructions hidden in the data it reads, the attack known as prompt injection.

The model came out that day in Anthropic's apps, in its API, and on the three major cloud platforms, at a price with which the company wanted to bring Opus models to more users. It also stood out at office tasks such as spreadsheets.

Sources

How to cite

“Claude Opus 4.5.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/claude-opus-4-5

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline