Detailed Analysis
Simon Willison released llm-anthropic 0.25 on April 16, 2026, a plugin update for his widely-used LLM command-line tool that significantly expands developer access to Anthropic's Claude model family. The centerpiece of the release is support for **Claude Opus 4.7**, Anthropic's most capable generally available model as of mid-2026, which excels in complex reasoning and agentic coding tasks relative to its predecessor, Claude Opus 4.6. The update introduces several technically meaningful additions: a `thinking_effort: xhigh` parameter for granular reasoning control, boolean options `thinking_display` and `thinking_adaptive` for customizing how the model's internal reasoning processes are surfaced, an increased default `max_tokens` ceiling aligned with the model's maximum output capacity of 128,000 tokens, and the removal of the now-obsolete `structured-outputs-2025-11-13` beta header for older models.
The practical significance of this release lies in how it lowers the barrier to accessing frontier-class AI capabilities through a familiar, terminal-native interface. Developers and researchers who rely on Simon Willison's LLM tool — a lightweight, extensible framework for interacting with language models from the command line — can now query Claude Opus 4.7 directly without navigating API documentation or constructing raw HTTP requests. The plugin's support for thinking-related parameters is particularly notable: by exposing `thinking_effort` and `thinking_adaptive` flags, the update gives practitioners direct control over the depth and style of the model's reasoning traces, a feature increasingly relevant as extended-thinking capabilities become standard in high-performance AI systems.
Claude Opus 4.7 itself represents a meaningful step in Anthropic's Claude 4.x series, which spans a tiered architecture of capability and speed. The series currently includes Claude Haiku 4.5 at the fast, cost-efficient end, Claude Sonnet 4.6 as the balanced middle tier with 64,000 output tokens and a May 2025 knowledge cutoff, and Claude Opus 4.7 at the high-capability apex with a 128,000-token output ceiling and a January 2026 knowledge cutoff. The model supports text and image inputs, multilingual tasks, and vision applications, and is accessible not only via the Claude API but also through Amazon Bedrock, Google Vertex AI, and Microsoft Foundry — reflecting Anthropic's expanding multi-cloud distribution strategy.
The llm-anthropic release fits into a broader trend of democratizing access to large language models through open-source tooling and plugin ecosystems. Willison's LLM project has become an important bridge between API-first AI providers and the developer communities that prefer scriptable, composable interfaces over graphical or web-based products. The simultaneous support across tools like LiteLLM — which also integrates Anthropic models with structured output and effort parameters for Sonnet 4.5/4.6 and Opus 4.1/4.5 — illustrates how the ecosystem around Anthropic's models has matured, with multiple third-party frameworks racing to keep pace with Anthropic's rapid model release cadence. As Claude models grow more capable and their parameter surfaces more sophisticated, plugins like llm-anthropic serve as critical translation layers, ensuring that cutting-edge features reach practitioners quickly and in ergonomic form.
Read original article →