Claude Opus
Anthropic's top-tier model. Opus 4.7 (2026) is the latest. Standard choice for hard reasoning, agentic tasks, code generation where quality > cost. ~$15/M input tokens, $75/M output. Slowest but most capable within Claude family.
Opus 4.7 features: Extended Thinking, native tool use, computer-use, 200k stable context. Benchmarks: SWE-bench Verified 80%+, MMLU 90%+, GPQA Diamond 88%+. Use cases: agentic coding (Cursor, Claude Code), complex research, mathematical proofs, multi-step planning. Anti-recommendation: customer-support chat, simple extraction tasks is overkill for Opus → use Haiku/Sonnet. Performance-price curve: Opus tasks that could have been done with Sonnet cost 5x for marginal quality gain. Modern best practice: route majority to Sonnet, escalate hard tasks to Opus.