Claude Opus 5 Is Here: Near-Frontier Intelligence at Half the Price
Anthropic released Claude Opus 5 on July 24, 2026, priced at $5 per million input tokens and $25 per million output, unchanged from Opus 4.8. The headline claim: intelligence close to Claude Fable 5, Anthropic's frontier model, at half the price. For GTM and RevOps teams the interesting part is not the coding leaderboards. It is that end-to-end business automation just got materially better without getting more expensive.
What did Anthropic announce?
Claude Opus 5 is available today on all platforms. It is the new default model on Claude Max, the strongest model on Claude Pro, and live on the API as claude-opus-5. A Fast mode runs at roughly 2.5 times default speed for twice the base price. Anthropic describes it as a step change for the Opus tier that powers long-running agents, with state-of-the-art results on coding and knowledge-work evaluations.
Two API betas shipped alongside the model: mid-conversation tool changes that no longer invalidate the prompt cache, and automatic fallbacks that route safety-flagged requests to another model instead of blocking them. Both matter if you run production agents.
How much does Claude Opus 5 cost?
$5 per million input tokens and $25 per million output tokens, exactly what Opus 4.8 cost. That pricing decision is the real story. If you already run Opus in production, the capability jump is free. If you run Fable 5 for quality reasons, Opus 5 now delivers close to the same results at half the price. Consistent with prior Opus models, there are no data retention requirements for general access.
What do the benchmarks say?
| Benchmark | Claude Opus 5 result |
|---|---|
| Frontier-Bench v0.1 | New state of the art; more than double the Opus 4.8 score at a lower cost per task |
| CursorBench | Within 0.5% of the Fable 5 peak score at half the cost per task |
| ARC-AGI 3 | Three times the score of the next-best model |
| Zapier AutomationBench | Around 1.5x the next-best pass rate at the same cost per task |
| OSWorld 2.0 (computer use) | Outperforms every model at any given cost; tops the Fable 5 best result at about a third of the cost |
| GDPval-AA (knowledge work) | New state of the art |
One caveat Anthropic states plainly: Opus 5 remains behind Mythos 5, the restricted-access frontier model, on cybersecurity, and its safeguards block exploit generation and penetration testing for general users.
Why this matters for GTM and RevOps teams
Three of those benchmarks measure business work, not code. Zapier's AutomationBench tests whether a model can finish a business task from start to finish, and Zapier reports Opus 5 took a raw account-health workbook and ran a complete churn-prevention sequence, flagging at-risk accounts, alerting the right owners, and summarizing for retention ops. Earlier models failed that task; Opus 5 passed it fully. That is not a developer demo. That is lifecycle marketing.
OSWorld measures whether a model can drive real software through the screen, the way a human operator would. Beating the previous frontier at a third of the cost moves computer-use agents from demo territory toward dependable ops tooling. And the Anthropic behavioral audit scored Opus 5 its most aligned model to date, with the lowest rate of reckless, hard-to-reverse actions. If you plan to let an agent touch a live CRM, that number matters more than any leaderboard.
Thinking about AI agents for your own stack? We build free personalized HubSpot Breeze starter packs, researched from your company's public footprint, in about 4 hours. Get your free starter pack |
Does this change anything for HubSpot Breeze users?
Not directly. Breeze agents abstract the underlying model away, so you do not pick Opus 5 inside HubSpot. Where model choice does show up is custom agents. HubSpot opened its Agent Hub and Agent Builder to public beta this week, and teams weighing a platform-native agent against a custom Claude build should read our Breeze vs custom Claude agent breakdown. If you want agents wired into your existing workflows instead, that is exactly what our workflow agents service covers. Cheaper near-frontier intelligence shifts the build-vs-buy math toward building, but only for teams with the data hygiene to support it.
Frequently asked questions
How much does Claude Opus 5 cost?
$5 per million input tokens and $25 per million output tokens, the same price as Opus 4.8. Fast mode, at roughly 2.5 times the speed, costs twice the base price.
Is Claude Opus 5 better than Claude Fable 5?
It comes close at half the price. Anthropic reports Opus 5 within 0.5% of the Fable 5 peak CursorBench score at half the cost per task, and ahead of Fable 5 on the OSWorld computer-use benchmark at about a third of the cost. Fable 5 remains the frontier model overall.
What is new for AI agents in this release?
Opus 5 topped the Zapier AutomationBench for end-to-end business tasks, and Anthropic shipped two API betas: mid-conversation tool changes without losing the prompt cache, and automatic fallbacks so flagged requests route to another model instead of failing.
Is Claude Opus 5 available on Claude Pro and Max plans?
Yes. It launched July 24, 2026 as the default model on Claude Max and the strongest model available on Claude Pro.
Source: Anthropic, Introducing Claude Opus 5 (July 24, 2026).
Founder of Hacking Demand. 12+ years building B2B demand generation, including 330+ B2B and B2C webinars produced.
Keep building demand
Want us to build your demand engine?
Done-for-you webinars, AI agents, and pipeline plays, live in days.
Talk to an expertBrowse the hacks