Claude Opus Glossary

The most-capable tier of Anthropic's *Claude* model family. The Dictionary's daily working model; the structural dependency named in *Opus Addict*.

Claude Opus is the most-capable tier in Anthropic’s Claude model family. The pricing and capability profile place Opus at the high end of the frontier-model market: best-in-class on most reasoning, writing, and agentic-tool-use benchmarks, with token costs an order of magnitude above the Sonnet tier. Major versions to date include Claude 3 Opus (March 2024), Claude 3.5 Opus (announced late 2024), Claude 4 Opus, 4.5 Opus, 4.6 Opus, and 4.7 Opus — the latter being the version Boris Cherny credits in his *Coding Solved * talk as the first model that “can hill-climb anything.”

For this Dictionary, Opus is the daily working model. Most entries in this corpus have been drafted in collaboration with Claude Opus, with the operator’s voice and editorial judgement as the final layer. The structural dependency the Dictionary names this as is Opus Addict — the recognition that most careful agentic work in 2026 depends on a single frontier closed-tier model, and the dependency has real downsides even when the model is good.

The token-cost profile of Opus drove the operator’s M5 Max purchase (see TOOLS.md cost analysis: cache writes accounted for 72.6% of Anthropic spend over a 16-day window in April 2026, with output tokens at only 4.8%) — and continues to shape the architecture choices documented in Sovereign Compute.

See also

Return to Dictionary All Entries (A–Z) For Students Other Writing