Anthropic releases Claude Opus 5 for everyday agent work

Anthropic has released Claude Opus 5, making the new model available today across its core Claude surfaces and the Claude API. The launch matters because Anthropic is not reserving this Opus-class jump for a narrow preview: it is positioning Opus 5 as the daily high-end model for coding agents, professional work, and longer sessions where users want stronger verification without moving all the way to Fable 5.

The company says Opus 5 comes close to Claude Fable 5 in many domains at half the price, while staying behind Mythos 5 on the most sensitive cybersecurity and biology risk areas. Independent coverage from The Verge confirms the release, the pricing frame, and Anthropic's emphasis on stronger cyber safeguards after the recent Fable 5 and Mythos 5 scrutiny.

Claude Opus 5 becomes the new Opus default

Anthropic says Opus 5 is now the default Opus model on Claude Max and the strongest model available on Claude Pro. Developers can call it through the Claude API as claude-opus-5, and the Claude Code changelog for July 24 lists Opus 5 as the default Opus model with a 1 million token context window.

That availability separates this from a lab-only benchmark announcement. Opus 5 is shipping into Claude.ai, Claude Code, Claude Cowork, and the platform API, with GitHub Copilot also listing Claude Opus 5 availability in its July 24 changelog. For teams already routing coding or analysis work through Claude, this is a practical model-selection event, not just a leaderboard update.

The pitch is Fable-like work at Opus pricing

Anthropic keeps Opus 5 at the same base API price as Opus 4.8: $5 per million input tokens and $25 per million output tokens. Fast mode is available at twice that base price on the Claude Platform, and through usage credits in Claude Code, for teams that prefer lower latency over cost.

The headline claim is not that Opus 5 replaces Fable 5. Anthropic still presents Fable 5 and Mythos 5 as stronger for the most ambitious long-horizon and dual-use frontier tasks. Opus 5 is pitched as the more efficient daily choice: stronger than Opus 4.8, close to Fable 5 on several coding and knowledge-work evaluations, and less expensive to run at scale.

For builders, the main evaluation detail is the shape of the gains. Anthropic says Opus 5 outperforms prior Opus models on Frontier-Bench, CursorBench, ARC-AGI 3, OSWorld 2.0, GDPval-AA, AutomationBench, HLE, and DeepSearchQA, with effort settings that let customers trade intelligence, speed, and token use. Treat those numbers as vendor-provided until more independent benchmark tables catch up, but the same release is already visible in first-party product notes and external reporting.

Cyber safeguards change routing behavior

The safety story is part of the release, not a footnote. Anthropic says Opus 5 does not advance the frontier in risky dual-use capability and remains behind Mythos 5 in biology research and offensive cybersecurity. At the same time, the company says it is more capable than Opus 4.8 at finding vulnerabilities, so it adds stronger guardrails on a narrow range of cyber tasks.

The operational change: flagged requests in Claude.ai, Claude Code, and Claude Cowork can fall back to Opus 4.8 by default, and API users can enable automatic fallbacks. Anthropic says those fallbacks let requests route to another model instead of simply blocking when Opus 5 or Fable 5 hits a safety classifier.

That matters for teams using Claude as an agent inside code review, security triage, and automation workflows. A model upgrade can otherwise create silent workflow breaks if a stricter safety layer rejects tasks that older models answered. Automatic fallback gives admins a cleaner migration path, but security teams still need to check whether a fallback model changes auditability, output quality, or policy expectations.

What teams should verify before switching

The fastest adoption path is a controlled routing test. Existing Claude users should compare Opus 5 against Opus 4.8 on their own long-running tasks, especially where browser use, code execution, document analysis, or multiple tool calls are involved. Claude Code users should also check their installed version because Opus 5 support appears in the July 24 2.1.219 changelog.

Teams that use Claude for regulated, security-sensitive, or bio-related work should read Anthropic's system card and verify fallback settings before broad rollout. Pricing is clear at the API level, but Fast mode, Claude Code usage credits, plan limits, and workspace policies can still change the real cost of a migration.

For readers comparing agent stacks, LinkLoot's guide to AI agent tools is the useful follow-up: this release is another sign that model choice is becoming a routing and governance problem, not a single dropdown preference.

Sources and methodology

This post uses Anthropic's release announcement as the primary source, cross-checks availability against Claude's public release notes and the Claude Code changelog, and uses The Verge for independent confirmation and context around pricing, safeguards, and government-testing pressure. OpenAI/Codex and Anthropic/Claude usage-limit/reset signals were checked during the same run; no universal new reset, doubled-limit window, or time-sensitive plan-entitlement alert was strong enough for a separate blog post.