What is agentic persistence in Claude Opus 5?

Agentic persistence is Claude Opus 5's signature trait: the model verifies its own work and iterates until a task actually succeeds, instead of stopping at a plausible-looking answer. Released July 24, 2026, it can run for extended periods, recover from errors, and navigate obstacles autonomously — including overnight coding sessions.

How much does Claude Opus 5 cost?

Claude Opus 5 is $5 per million input tokens and $25 per million output tokens — the same price as Opus 4.8, but at roughly half the cost of Claude Fable 5. It has a 1M-token context window and 128K maximum output, and is now the default Opus on Claude Max.

Quick Answer

What Is Claude Opus 5 Agentic Persistence? (2026)

Q: Do I still need to prompt Claude Opus 5 to verify its work?

No. Because Opus 5 self-verifies without instruction, Anthropic recommends removing prompts like 'include a final verification step' or 're-verify before responding.' Keeping them causes over-verification and wasted tokens without improving output quality.

Published: August 3, 2026

The Short Answer

Agentic persistence is Claude Opus 5’s signature behavior: it verifies its own work and keeps iterating until a task actually succeeds, rather than stopping at a plausible answer. Released July 24, 2026, Opus 5 self-corrects, recovers from errors, and can run long autonomous sessions.

What Self-Verify Actually Does

Opus 5 performs self-verification without being told to, including improved root-cause debugging. It identifies and fixes its own mistakes without extra prompting. Concrete examples from Anthropic’s evaluations:

Asked to rebuild a 3D model from a drawing it couldn’t directly view, Opus 5 built its own computer-vision pipeline to extract geometry from pixels.
When a live data feed was unavailable, it built its own test harness to validate code.
In production, it manages its own memory for more autonomous monitoring agents.

Why Old Verification Prompts Now Hurt

Because self-verification is built in, prompts like “include a final verification step” or “re-verify before responding” are no longer necessary. Anthropic warns they cause over-verification and wasted tokens without improving quality. Opus 5 also narrates its progress and corrections more frequently during agentic sessions.

Key Facts

Spec	Claude Opus 5
Released	July 24, 2026
Context	1M tokens
Max output	128K tokens
Price	$5 / $25 per MTok
Position	Default Opus on Max; ~half the price of Fable 5

Why It Matters

Agentic persistence targets the biggest failure mode of coding agents — stopping too early. For overnight refactors, long bug hunts, and long-horizon tasks, a model that keeps going until tests pass is a meaningful shift.

Sources

Anthropic — Claude Opus 5: anthropic.com/news/claude-opus-5
Anthropic — Prompting Claude Opus 5: platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5
AWS — Claude Opus 5 on Bedrock: aws.amazon.com/blogs/machine-learning/introducing-claude-opus-5-on-aws