Anthropic launches Claude Opus 5 at Opus 4.8 pricing

Anthropic launched Claude Opus 5 on July 24, 2026, available immediately across all platforms. It calls the model "thoughtful and proactive," coming close to Fable 5's frontier intelligence at half the price, and its biggest Opus jump since 4.5.

Key points

• Performance: state-of-the-art on evaluations like Frontier-Bench v0.1 (beating every model, more than doubling Opus 4.8 at a lower cost per task) and GDPval-AA; on CursorBench 3.2 it lands within 0.5% of Fable 5 at half the cost. ARC-AGI 3 score is ~3x the next-best model; OSWorld 2.0 (computer use) beats all models at any given cost.
• Pricing: unchanged from Opus 4.8 - $5 per million input tokens and $25 per million output tokens. A Fast mode runs ~2.5x default speed at twice the price. API id: claude-opus-5. It is the new default on Claude Max and the strongest model on Claude Pro.
• Safety & alignment: Anthropic says it is its most aligned model to date, with the lowest misaligned-behavior score (2.3). On dangerous capabilities Opus 5 stays behind Mythos 5 in biology and offensive cyber; it nears Mythos 5 at finding vulnerabilities but lags far behind at exploiting them.
• Cyber safeguards: cyber classifiers are less restrictive than Fable 5's (~85% fewer interventions), allowing source-code vulnerability finding while blocking binary scanning, pen-testing and exploit generation; blocked requests fall back to Opus 4.8.
• For developers: two betas ship alongside - mid-conversation tool changes without invalidating the prompt cache, and automatic API fallbacks so flagged requests reroute to another model instead of being refused.

FAQ

Is Opus 5 more expensive than Opus 4.8? No - same $5 in / $25 out per million tokens, but materially higher performance, so cost per task is often lower. Opus 5 or Fable 5? Anthropic positions Opus 5 as an everyday model: close to Fable 5 on many coding/agent tasks but far cheaper; Fable 5 still edges peak intelligence, while Mythos 5 leads on biology and exploit development. Where can I use it? On the Claude API (claude-opus-5), Amazon Bedrock, Google Cloud, Microsoft Foundry and GitHub Copilot. Full benchmarks and the System Card are in the original announcement - see Read original.