Detailed Analysis
Claude Code users seeking to restore or explicitly set claude-opus-4-6 as their default model — particularly with the 1 million token context window — can do so by exporting the `ANTHROPIC_MODEL` environment variable directly in their shell configuration file. The community-shared method involves appending an export command to `.zshrc` (or the equivalent Windows shell profile) and restarting the CLI, with the `[1m]` suffix optionally added to invoke the extended context variant. The linked official Anthropic support documentation at `support.claude.com` confirms this environment variable approach as the sanctioned configuration path, and Claude Code is designed to strip the `[1m]` suffix before making the underlying API call, meaning users need not worry about the suffix causing errors at the request level.
The need for this configuration tip reflects a meaningful shift in how Anthropic is managing model defaults across its subscription tiers. As of April 2026, Max and Team Premium subscribers default to Opus 4.7, while Pro, Team Standard, Enterprise, and direct API users default to Sonnet 4.6 — with a further migration to Opus 4.7 scheduled for April 23, 2026. Bedrock, Vertex, and Foundry deployments default to Sonnet 4.5. Claude Code also applies an automatic fallback to Sonnet when Opus usage hits defined thresholds, meaning users who wish to consistently operate on Opus 4.6 must actively anchor that preference through environment-level configuration rather than relying on platform defaults.
The 1 million token context window in Opus 4.6 represents a significant capability milestone. Generally available since March 13, 2026, and priced at standard rates for Max, Team, and Enterprise subscribers, the 1M context support positions Opus 4.6 as the first Opus-class model to reach this scale. The model supports up to 128,000 output tokens and introduces extended thinking through an `effort` parameter — defaulting to `high` on Opus 4.6 for most plans — which replaces the previously deprecated `thinking` parameters. Users working on complex, long-horizon coding tasks stand to benefit substantially, as a 1M context window allows entire large codebases, documentation sets, or multi-file projects to be held in a single context without truncation.
Broader trends in AI tooling are visible here: the proliferation of model variants, context tiers, and subscription-dependent defaults is increasingly pushing operational complexity onto end users. The fact that a community post is necessary to clarify how to "go back" to a prior model — even one still officially supported — illustrates how rapidly the model landscape is evolving and how default-switching by providers can disrupt established workflows. Developers who have optimized prompts, effort levels, or agentic pipelines around Opus 4.6's specific behavior profile have a legitimate interest in pinning that model explicitly, independent of what the platform chooses as its rolling default. Anthropic's support for per-model environment variable overrides and server-managed settings fields offers a reasonable escape valve, but only for users who know to look for it.
Read original article →