Artificial Intelligence

Opus 5 Costs Half of Fable 5, If You Set It Right

Anthropic released Claude Opus 5 on 24 July at 5 dollars per million input tokens, unchanged from Opus 4.8. The pricing dial, not the discount, is what changes your budget.

Artificial IntelligenceBy Servola Tech Desk2026-07-243 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Opus 5 Costs Half of Fable 5, If You Set It Right

A frontier model at last year's Opus price

Anthropic put Claude Opus 5 on every one of its platforms on 24 July, and the number that matters is the one that did not change. Input costs 5 dollars per million tokens and output costs 25 dollars, identical to Opus 4.8, while Fable 5, the company's flagship since June, lists at 10 and 50. On Frontier-Bench v0.1 Opus 5 scored 43.3 percent against Opus 4.8's 18.7 percent and Fable 5's 33.7 percent. On CursorBench 3.2 it came within half a percentage point of Fable 5's peak, and on OSWorld 2.0, a computer-use test, it beat Fable 5's best result at just over a third of the cost.

The model is live on the API as claude-opus-5, in Claude Code, in Claude Cowork and on Claude.ai, where it is now the default on Max and the strongest option on Pro. Anthropic also reports gains across every life-sciences evaluation it tracks, with more than ten percentage points on organic chemistry. For an operator, the headline is simple enough: the work you were sending to the expensive tier can go to the cheaper one without a capability argument.

Fast mode costs what Fable 5 costs

Here is the part the launch coverage skipped. Fast mode bills at twice the base rate, which means Opus 5 running fast costs 10 dollars per million input and 50 per million output. That is not a discount on Fable 5. That is Fable 5's list price, charged for a model that Anthropic itself positions one notch below it. The saving everyone is repeating exists only at the base setting, and it disappears the moment a default flips.

This is the structural change worth naming. A model used to be a fixed unit cost you chose during architecture review. Opus 5 ships with a capability-versus-cost toggle, and Anthropic's own benchmark notes lean on it, pointing out that even the cheapest effort level clears rivals' best scores on Zapier's AutomationBench. That is a genuine engineering gift and a genuine governance problem, because the cost of the identical job now varies by a factor of two or three depending on a setting that any engineer can change in a config file.

Recalculate before your next invoice

Run your own arithmetic rather than the vendor's. Take a modest production agent that reads 50 million input tokens and writes 5 million output tokens a month. On Opus 5 at base rates that is 250 dollars plus 125 dollars, so 375 dollars a month. The same workload on Fable 5 is 500 plus 250, so 750 dollars. Switch that Opus 5 agent to fast mode and you are back at 750. The migration saves you half, or nothing, and the deciding variable is a flag, not a contract.

So the action this week is narrow. Find every place an effort or speed setting is set, write down the default, and decide who owns changing it. For a European team there is a second line to add: these are dollar prices, so your euro or sterling cost moves with the exchange rate as well as with the setting, and a budget signed off in euros against a dollar rate card needs a stated assumption. Ask your finance lead which rate the forecast used before you promise a saving.

Frequently asked questions

Is Claude Opus 5 cheaper than the model it replaces?

No. It costs the same as Opus 4.8, at 5 dollars per million input tokens and 25 per million output. What changed is capability at that price. It is cheaper than Fable 5, which lists at 10 and 50, so moving work down from Fable 5 to Opus 5 is where the saving comes from.

What does fast mode do to the bill?

Fast mode bills at twice the base rate, so 10 dollars per million input and 50 per million output. That matches Fable 5's list price. If your agents run in fast mode by default, the price advantage over Fable 5 is gone.

Should we migrate our agents to Opus 5 now?

Test it on your own workload rather than trusting the benchmarks, because published scores are measured on tasks that are not yours. If it holds up, the move down from a flagship tier is the saving worth taking, but fix the effort-setting defaults and their ownership first, or the saving will not survive the quarter.

The frontier is no longer the scarce thing; deciding how much of it to buy per task is. The teams that write that decision down, and name who can change it, will be the ones whose AI budget still means something in October.

AnthropicClaude Opus 5Claude Fable 5LLM pricingAI agentstoken costsinference budgets

Independent Tests Rank Kimi K3 Above Fable 5

On an independent blind leaderboard of 1757 developer votes, Moonshot's open Kimi K3 ranked first for frontend code, ahead of Claude Fable 5, even though its own maker placed it second overall.

4 min read

Artificial Intelligence

OpenAI Just Halved the Price of Everyday AI

GPT-5.6 shipped July 9 in three tiers. Terra matches last year's GPT-5.5 at half the price, and Luna sets a new floor. The owner move is to reprice and route by task tier.

3 min read

Law & Regulation

The EA Buyout Now Hinges on Brussels

The largest take-private in history has cleared EA shareholders and US antitrust. The last real gate is the EU - and not classic merger review, but the Foreign Subsidies Regulation, facing its first blockbuster test.