Claude Opus 5 pricing keeps Opus cost as Anthropic targets daily AI work

Anthropic released Claude Opus 5 on Friday, with Claude Opus 5 pricing set at $5 per million input tokens and $25 per million output tokens, the same rates as Opus 4.8. The company is positioning the model as a lower-cost workhorse for coding, agents and enterprise workflows, while saying its Fable 5 model remains the top option for the most demanding long-running tasks.

Opus 5 is available now across Anthropic’s platforms. It becomes the default model for Claude Max, the company’s premium consumer plan, and the strongest model offered on Claude Pro. Anthropic did not say that Opus 5 is its most capable system overall, and acknowledged that rival models remain ahead in some areas.

How much does Claude Opus 5 cost?

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens. Anthropic is also offering a Fast mode that runs at roughly 2.5 times the default speed for twice the base price, according to the company.

The pricing is the main commercial point. Anthropic says Opus 5 provides near-Fable performance for many bounded tasks at lower cost, a pitch aimed at customers whose AI budgets are increasingly tied to inference volume rather than pilots or demos.

What Anthropic says Opus 5 can do

Anthropic says Opus 5 sets new marks on several coding and knowledge-work evaluations. On Frontier-Bench v0.1, an agentic terminal coding test, the company reports a score of 43.3%, compared with 18.7% for Opus 4.8 and 33.7% for Fable 5. Anthropic also says Opus 5 scored three times higher than the next best model on ARC-AGI 3, and beat Fable 5’s best OSWorld 2.0 result at just over one-third of the cost.

The caveats matter. Anthropic says Opus 5 still trails Mythos 5 on cybersecurity and biology research, and that an OpenAI-family model leads on one agentic coding benchmark. Anthropic also distinguishes Opus 5 from Fable 5 by task type: Opus 5 is presented as stronger for bounded work with clear outcomes, while Fable 5 is meant for longer autonomous jobs that require coherence across many steps over hours or days.

Several customers cited efficiency rather than raw benchmark gains. Harvey said Opus 5 matched Opus 4.8’s maximum-reasoning performance while producing 26% fewer tokens on average, according to Niko Grupen, its head of applied research. Richard Pham of Fundamental Research Lab said the model improved accuracy by nine percentage points on difficult financial-modeling tasks while using about one-third fewer turns and tool calls and 60% less time. Zapier CEO Wade Foster said Opus 5 completed a churn-prevention workflow on the company’s AutomationBench that previous models did not pass.

Safety, fallbacks and enterprise friction

Anthropic says its automated behavioral audit rated Opus 5 as its most aligned model so far, with an overall misaligned-behavior score of 2.3. The company also says it did not train Opus 5 on cyber tasks, although general capability gains still improved its performance there.

On Anthropic’s OSS-Fuzz evaluation, Opus 5 found vulnerabilities at a 79.4% rate, close to Mythos 5’s 80%, but succeeded at creating exploits in 4 challenges versus 13 for Mythos 5. Anthropic says cyber classifiers should intervene about 85% less often for Opus 5 than for Fable 5. When a classifier is triggered in Claude.ai, Claude Code or Claude Cowork, requests fall back to Opus 4.8 by default, with a visible notice to users.

The launch comes as Anthropic is under pressure to turn high usage into durable revenue. Reuters reported in February that Anthropic was valued at roughly $380 billion in its latest funding round. Contrary Research estimated Claude had about 40% of enterprise large-language-model usage in late 2025 and said Claude Code had reached about $1 billion in annualized revenue.

Anthropic also faces large infrastructure commitments, including a reported $30 billion Azure compute deal, plus arrangements with Google Cloud and Nvidia. Against that backdrop, holding Opus pricing flat while improving performance per task is effectively a bid to make more enterprise automation workloads economical enough to run repeatedly.

This story draws on original reporting from VentureBeat.