Anthropic Claude Opus 4.8 — strongest coding scores yet

(released May 28, rolling out this week) — 88.6% on SWE-bench Verified and 74.6% on Terminal-Bench 2.1, with parallel-subagent workflows and a 2.5x fast mode, all at the same $5/$25 per-1M pricing. Why it matters: top-tier agentic coding performance with no price increase — relevant if you're picking a default model for autonomous coding tasks.

---

Sources: LLM-Stats deep dive