Anthropic released Claude Opus 5 on July 24, making it available as the default model on Claude Max and the strongest model on Claude Pro. The company states that Opus 5 comes close to Claude Fable 5's frontier intelligence at half the price, while CNBC reports pricing of $5 per million input tokens and $25 per million output tokens. Anthropic also reported that the model is not state-of-the-art for cybersecurity-related dual-use capabilities.
Anthropic released Claude Opus 5 on July 24. The company says the model is available immediately, comes close to Claude Fable 5's frontier intelligence at half the price, is the default model on Claude Max, and is the strongest model offered through Claude Pro.
CNBC reports that the model costs $5 per million input tokens and $25 per million output tokens, or half the price of Claude Fable 5. Anthropic states that Opus 5 delivers improved performance at the same cost as its predecessor, Claude Opus 4.8.
Benchmark claims and effort settings
Anthropic reports that Opus 5 outperforms other models on its internal Frontier-Bench v0.1 evaluation, using the mini-SWE-agent harness and a Google Kubernetes Engine backend. The company says the model more than doubles Opus 4.8's performance at a lower cost per task on that benchmark.
On CursorBench 3.2, Anthropic reports that Opus 5 at maximum effort scored within 0.5% of Fable 5's peak score while costing half as much per task. The company also reports stronger performance at comparable cost across high, xhigh, and maximum effort settings.
Anthropic states that customers can select effort settings to trade off intelligence, token consumption, speed, and cost. This type of runtime control is increasingly relevant for engineering teams operating agentic coding or workflow systems, where model quality alone does not determine production economics.
Capability boundaries and enterprise economics
Anthropic reports that Opus 5 is state-of-the-art on coding and knowledge-work evaluations, but not on cybersecurity tasks involving risky dual-use capabilities, where it says Claude Mythos 5 remains ahead. The distinction matters because benchmark leadership can vary substantially by task domain, tool-use harness, and inference budget.
CNBC describes the launch as arriving amid greater enterprise scrutiny of AI return on investment and competition from lower-cost model providers. Dianne Penn, Anthropic's head of product management for research, told CNBC: "Enterprises, in our feedback and with our customer base, are looking for value." She added that a cheaper offering is not useful if it cannot deliver similar quality.
For ML teams, the published comparison reinforces a familiar deployment pattern: model selection increasingly requires measuring task-level success rates, latency, and token cost together, rather than choosing solely on a single aggregate benchmark.
Key Points #
- 1Anthropic released Opus 5 with lower token prices, pairing frontier coding claims with a cost-focused deployment proposition.
- 2Company-reported benchmark results show effort settings can materially affect quality and cost, making workload-level evaluation important for teams.
- 3Enterprise AI buyers increasingly assess return on investment, a market pattern that rewards comparable task quality at lower inference cost.
Scoring Rationale #
Claude Opus 5 is a major frontier-model release with reported gains in coding and knowledge-work performance alongside materially lower token pricing. Its effort settings and cost-per-task claims are directly relevant to teams running production AI agents and evaluating inference economics.
Sources #
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.