Claude Opus 5

Overview

Anthropic’s July 24, 2026 Opus-class model positioned near Claude Fable 5 capability at half the price ($5/$25 per MTok). API ID claude-opus-5. Default on Claude Max; strongest on Claude Pro.

Info

Bench SOTA and alignment rankings are Anthropic claims unless independently reproduced.

Recent Developments

• 2026-07-24: arc-prize verifies 30.16–30.2% on arc-agi-3 (High only) — ~4× prior gpt-5-6-sol ~7.8% Max; five new Public Demo clears (2026-07-26-arc-prize-claude-opus-5-results)

• 2026-07-24: Public launch — SOTA claims on Frontier-Bench; strong CursorBench/AutomationBench/OSWorld; Fast mode + API fallbacks (2026-07-24-anthropic-claude-opus-5-launch)

• Safety: company says most aligned Opus to date; cyber classifiers ~85% fewer interventions than Fable 5; behind Mythos 5 on exploit/biology dual-use

Related

• anthropic-claude-opus-5-launch

• automation-bench

• osworld

• anthropic

• claude

• claude-fable-5

• claude-opus-4-8

• frontier-bench

• llm-pricing-parity

• coding-agents

Sources

• 2026-07-24-anthropic-claude-opus-5-launch
• 2026-07-24-anthropic-claude-opus-5-cnbc
• 2026-07-24-anthropic-claude-opus-5-venturebeat