How much does Claude Opus 5 cost?

Five dollars per million input tokens and 25 dollars per million output tokens, half of Fable 5's price and exactly what its predecessor, Opus 4.8, cost. The model is also notably token-efficient, so the per-task cost is low.

Is Claude Opus 5 available on subscription plans?

Yes, on every platform: in the Claude app on Pro, Max, Team and Enterprise plans (as the new default on Max), plus the API, Claude Cowork and Claude Code. Unlike Fable 5, it is included in the regular Pro plan.

What changed compared to Opus 4.8?

An order-of-magnitude leap at an unchanged price: on novel problem solving its score jumped from 1.5 to 30 percent. The most important qualitative change is self-verification: the model keeps iterating and checking its own work until the task is genuinely done.

Claude Opus 5 is here: top-tier performance at half the cost

Q: Is Claude Opus 5 better than Fable 5?

On several evaluations, yes: it leads agentic terminal coding, knowledge work, agentic search and business workflows at half the cost. At maximum effort Fable 5's peak score stays ahead, and Fable remains stronger on legal and health tasks.

AI news 24 July 2026 · dropmind editorial

On 24 July 2026 Anthropic released Claude Opus 5, and with it reshuffled its own model line-up: the new Opus beats the flagship Fable 5 on several coding and knowledge-work evaluations, costs half as much, and, unlike Fable, is included in the regular Pro plan.

Released24 July 2026

Price$5 / $25 per M tokens

AccessClaude app · API · Cowork · Code

Positionthe strongest Opus

The number five laid out from speckled egg-like pebbles beside a coral starburst on a cyan-mint gradient: Claude Opus 5 launch visual

Fable-level intelligence at half the price

According to Anthropic's announcement, Opus 5 is the company's best price-to-performance model yet: nearly all of Fable 5's intelligence at roughly half the cost per task. Pricing is unchanged from Opus 4.8, 5 dollars per million input tokens and 25 dollars per million output tokens, while performance has doubled in several areas. The model also exposes effort levels: you choose how hard it should think about a task, tuning the balance between cost and capability per job.

The availability is just as big a story: Opus 5 ships immediately on every platform, in the Claude app from Pro upwards (as the new default on Max), via the API, in Claude Cowork and in Claude Code. Unlike Fable 5 (our article on it is in Hungarian), which draws on separate usage credits on the Pro plan, Opus 5 is part of the regular subscription: near-flagship intelligence has become the everyday baseline for the first time.

What the evaluations show

The published results are unusually decisive. On agentic terminal coding Opus 5 scores 43 percent against Fable 5's 33 and Opus 4.8's 21. On novel problem solving the jump is an order of magnitude: from Opus 4.8's 1.5 percent to 30. On knowledge work, agentic search, computer use and business workflows it leads every other frontier model, GPT‑5.6 Sol included, and it tops Zapier's automation benchmark at every effort level.

For the full picture: at maximum effort Fable 5's peak score stays in front, with Opus 5 close behind at half the cost per task. And two domains clearly remain Fable's: legal and health work. Early hands-on impressions, though, are consistent: for everyday knowledge work the two models' outputs are on par, and Opus is often faster.

Self-verification is the real leap

More important than any benchmark is a qualitative change: Opus 5 is much stronger at checking its own work, iterating until the result actually meets the goal. That is the key capability behind agentic, multi-step workflows, because the quality of a long task chain depends on whether the model catches and fixes its own mistakes. In one published example it was asked to rebuild a machine part as a 3D model from a drawing it had no way to view: it responded by writing its own computer-vision pipeline, extracting the geometry from raw pixels and reconstructing the part from that.

Visual output has improved as well (the demos include a working wind tunnel simulation and an interactive 3D diagram), misaligned behaviour is rarer, and the safety classifiers trigger far less often than on Fable 5, which means fewer needless interruptions in day-to-day work.

What does this mean for your business?

The practical formula just got simpler. Until now a well-sized system meant cheaper models carrying the routine and an expensive flagship reserved for the hardest problems. Opus 5 moves that line up: the everyday default can now be near-flagship intelligence, at half the API price, and at no extra cost on subscription. If you run Claude-based systems, it is worth reviewing where the pricier model can be swapped for Opus 5, typically with no loss of quality, and where Fable 5 still earns its keep, such as maximum-effort, legal or health-adjacent tasks.

Curious how much a well-chosen model swap would save in your own systems? Let's find out together on a free 30-minute consultation.

Free 30-minute consultation

Frequently asked questions

How much does Opus 5 cost?

Five dollars per million input tokens and 25 dollars per million output tokens, half of Fable 5's price and exactly what Opus 4.8 cost. It is also token-efficient, so per-task cost is low.

Is it available on subscription?

Yes, on every platform: in the Claude app from Pro upwards (default on Max), via the API, in Claude Cowork and Claude Code. Unlike Fable 5, it is included in the regular Pro plan.

Is it better than Fable 5?

On several evaluations, yes: terminal coding, knowledge work, agentic search and business workflows, at half the cost. At maximum effort, and on legal and health tasks, Fable stays ahead.

What changed from Opus 4.8?

An order-of-magnitude leap at an unchanged price, and the headline change is self-verification: the model keeps iterating until the task is genuinely done.

Sources

Last updated: 24 July 2026