Claude Opus 4.6 brings one million token context window to Anthropic's flagship model

2026-02-05 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Cybersecurity & Data Privacy · Depth: Intermediate, short

Summary

Anthropic has released Claude Opus 4.6, its new flagship large language model, featuring a one-million-token context window now available in beta. This model significantly improves retrieval accuracy, scoring 76 percent on the MRCR v2 test with one million tokens, compared to its predecessor Sonnet 4.5's 18.5 percent. Opus 4.6 also leads in knowledge work benchmarks, achieving 1,606 Elo points on GDPval-AA, surpassing OpenAI's GPT-5.2 by 144 points. New API features include "Adaptive Thinking" and "Compaction" for managing context, alongside increased maximum output to 128,000 tokens. Despite performance gains, the model exhibits a slight increase in vulnerability to indirect prompt injection attacks.

Key takeaway

For CTOs and VPs of Engineering evaluating large language models for complex document processing or knowledge work, Claude Opus 4.6 presents a compelling option due to its one-million-token context window and superior benchmark performance in retrieval and knowledge tasks. However, you should carefully assess its increased vulnerability to indirect prompt injection, especially for agentic AI applications, and implement robust security measures.

Key insights

Claude Opus 4.6 offers a one-million-token context window with enhanced retrieval and knowledge work performance.

Principles

• Larger context windows can degrade performance without mitigation.
• Benchmarking provides a rough indicator of real-world performance.

Method

Anthropic addresses "context rot" with model improvements and a "Compaction" feature that summarizes older context automatically.

In practice

• Use "Compaction" to manage large conversation histories.
• Adjust the "effort parameter" for simpler tasks to reduce costs.
• Explore "Agent Teams" for parallel AI agent task execution.

Topics

• Claude Opus 4.6
• Large Context Windows
• LLM Benchmarks
• Prompt Injection
• API Integrations

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, Data Scientist

Related on AIssential

• Anthropic rolls out Claude Opus 4.6 for long-context workloads — Claude Opus 4.6 significantly expands context and improves agentic coding, setting new benchmarks in multidisciplinary reasoning and knowledge work.
• Opus 4.6 is here — Claude Opus 4.6 offers enhanced planning, sustained task execution, and superior code debugging with a 1M token context.
• Claude Opus 4.6 in 8 mins! — Opus 4.6 offers a 1M context window and advanced agentic capabilities, but at a premium cost.
• Anthropic just dropped Opus 4.6... — Claude Opus 4.6 significantly advances LLM agentic autonomy and context handling, especially for complex coding and enterprise tasks.
• Anthropic's Claude Opus 4.7 — Claude Opus 4.7 enhances agentic coding, computer vision, and knowledge work, but introduces tokenization and long-context performance tradeoffs.
• Anthropic's Claude Opus 4.6 brings 1M token context and 'agent teams' to take on OpenAI's Codex — Anthropic's Claude Opus 4.6 advances AI with a 1M token context and "agent teams," intensifying enterprise competition.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.