Anthropic releases Claude Opus 4.6 with agent teams

Background

On 5 February 2026 Anthropic released Claude Opus 4.6, which it described as its smartest model at the time. ^[1]

What happened

Opus 4.6 was the first Opus-class model with a 1M-token context window, which is the amount of text a model can consider in a single request. The window was offered in beta, and the model supports up to 128K output tokens. ^[1] Pricing stayed at $5 per million input tokens and $25 per million output tokens. ^[1]

Key details

Claude Code gained agent teams, a research preview in which multiple agents work in parallel and coordinate with one another. ^[1][2] The release also introduced adaptive thinking, four effort levels, and context compaction, which was offered in beta. Context compaction shortens the earlier part of a long session so that work can continue within the context limit. ^[1]

Claude in PowerPoint launched as a research preview for Max, Team and Enterprise plans, and Claude in Excel was upgraded at the same time. ^[1]

Anthropic reported leading results on Terminal-Bench 2.0, on Humanity's Last Exam, where Opus 4.6 scored 53.0% with tools, and on GDPval-AA. ^[1]

Several terms in this release need a short explanation. A research preview is an early version of a feature offered to a limited group so that the company can learn from its use, and it may change before general release. A beta is a version available for testing before it is considered stable. Adaptive thinking lets the model decide how much reasoning a given request needs, and the effort level sets how much reasoning it is asked to use.

Impact

Agent teams were offered as a research preview, and the 1M-token context window was in beta at release. ^[1][2]