Anthropic ships Claude Opus 5

2026-07-24 — Friday

Anthropic released Claude Opus 5 today. The framing in the launch post is unusually specific: a “thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price”. It arrives at the same price as Opus 4.8 — $5 per million input tokens, $25 per million output — which makes this a capability jump with no bill attached.

Coding and software engineering

• Frontier-Bench v0.1 — more than doubles Opus 4.8’s performance, at a lower cost per task.
• CursorBench 3.2 — within 0.5% of Fable 5’s peak, at half the cost.
• AA Coding Agent Index — the best performance at a given cost across effort levels.

Knowledge work and reasoning

• ARC-AGI 3 — a score three times the next-best model.
• Zapier AutomationBench — a pass rate around 1.5× the next-best model.
• OSWorld 2.0 — surpasses Fable 5 at just over a third of the cost.

On scientific work, Anthropic reports organic chemistry up 10.2 percentage points on Opus 4.8 and protein-related tasks up 7.7 points.

What it means in practice

Opus 5 carries a 1M-token context window and 128K maximum output, with adaptive thinking on by default — a change from Opus 4.8 and 4.7, where omitting the thinking parameter meant no thinking at all. Fast mode runs roughly 2.5× the default output speed.

On alignment, Anthropic reports an automated behavioural audit score of 2.3 on misaligned behaviour, the lowest of its recent models, while noting Opus 5 remains behind Mythos 5 on cybersecurity exploitation tasks.

Featured Videos

Sources

• Anthropic — Introducing Claude Opus 5
• TechCrunch — Anthropic launches Opus 5
• InfoWorld — Anthropic releases ‘more efficient’ Claude Opus 5
• Claude Timeline — Model release dates and version history