Detailed Analysis
Anthropic's Claude Code has quietly rolled out a significant default configuration change for its Opus 4.7 model, enabling a 1 million token context window as the standard setting for all users. The shift was noticed organically by users who observed unusually low context window usage percentages on their first prompts of a new session — a counterintuitive signal that revealed the denominator had grown dramatically. Confirmation via the `/context` command showed the 1M context window active where it previously had not been. Alongside this expansion, Opus 4.7 supports up to 128,000 output tokens and includes an adaptive thinking capability, though the latter remains off by default and must be explicitly enabled via `thinking: {type: "adaptive"}`. Notably, thinking content is also now omitted from responses by default, requiring an opt-in `display: "summarized"` configuration to restore visibility into the model's reasoning process.
The practical implications for software development workflows are substantial. A 1 million token context window can accommodate roughly 30,000 lines of code in a single prompt, enabling developers to load entire microservice architectures, full framework documentation, or complete SaaS application codebases without chunking or retrieval workarounds. This removes one of the most persistent friction points in AI-assisted development: the need to carefully curate what context is fed to the model at any given time. Anthropic has positioned this expansion competitively by offering it at standard API pricing with no long-context premium — tokens beyond the 200K threshold are billed at the same rate as earlier tokens, a notable departure from the surcharge models some competitors have employed.
The accuracy implications are equally significant. Research cited in the release documentation shows Opus 4.7 achieving 76% accuracy on retrieval tasks across million-token contexts, compared to just 18.5% for Sonnet 4.5 — a gap that underscores the importance of model architecture in actually leveraging large context windows effectively. Expanding the window without maintaining retrieval fidelity would offer little real-world benefit, making this accuracy figure a critical validation of the feature's utility. The fact that Anthropic has made 1M context the *default* rather than an opt-in configuration signals confidence that the model can handle large contexts without degraded performance in typical use cases.
This development fits into a broader industry trajectory in which context length has become a primary competitive dimension for frontier AI models. Google's Gemini models have long marketed multi-million-token windows, and the race to offer expansive context at accessible price points has accelerated through 2025 and into 2026. Anthropic's move to make 1M context the default in Claude Code — rather than requiring users to configure it manually — reflects a maturation of the technology from novelty to expected infrastructure. It also acknowledges that developer workflows increasingly demand persistent, deep awareness of large codebases rather than the conversational, stateless interactions for which earlier context limits were designed. The option to disable the 1M window, documented in community posts, further reflects Anthropic's acknowledgment that some users may prefer leaner sessions for performance or cost-management reasons, preserving flexibility while shifting the default toward maximum capability.
Read original article →