Opus 5 Overtook Anthropic's Own Flagship in a Month. That Gap Is a New 'AI Spend Routing Audit' Service Business.

Anthropic just proved, with its own product line, that most companies are not paying for the best model.

They are paying for whichever model survives a cost decision.

Ramp's AI Index for July 2026 found that Fable 5, Anthropic's flagship, "supplied 6% of the Anthropic tokens" Ramp tracked but accounted for "11.4% of attributed Anthropic spending" — a plateau, not a takeover.

Then Implicator.ai, citing Ramp and the Financial Times, reported the real story: "Within a month of its launch, Anthropic's cheaper Opus 5 had overtaken its flagship Fable 5 in corporate spending."

Opus 5 launched on July 24 priced at "$5 per million" input tokens and "$25 per million" output tokens — "half of Fable's rates." Anthropic itself is now positioning the two models explicitly: "Opus 5 for value and Fable 5 for 'days-long, very autonomous projects.'"

The business idea

Most companies buying AI at scale have never deliberately set up cost-based routing. They picked a model when they started building, and everything still runs through it — premium and routine tasks alike.

That is the audit.

You sell a short, paid engagement that answers one question: which of this team's AI tasks actually need the expensive model, and which ones are burning premium-tier spend on routine work?

Implicator.ai's framing of the underlying strategy is the pitch, almost word for word: "A company can send routine work to a cheaper system and reserve a costly one for assignments that fail elsewhere or require sustained autonomy."

Why this works now

Because the market leader just modeled the exact behavior you are selling.

Anthropic's own customers didn't wait to be told to route by cost. Vercel's AI Gateway data (also cited by Implicator.ai) showed Fable pulling in a wave of new demand — "nine in 10 Fable teams were new users" — alongside a "59% token volume increase," a "37% spending increase," and a "13.6% price decrease" across the gateway in the same window. Usage is growing and prices are falling at the same time, which means the routing decision only gets more valuable to get right, not less.

Ramp also found that "43.5% of U.S. businesses use Anthropic products" against "39.7%" for OpenAI, in its sample — meaning most of your addressable market for this audit is already running multi-model AI spend today, not hypothetically.

Best customer profile

This works best for teams that:

• already have a real AI bill, not a pilot budget
• run recurring workloads through one default model regardless of task complexity
• have engineering capacity but no one dedicated to watching model pricing changes
• are worried about a surprise spend spike the next time a lab reprices

Good examples: AI-feature SaaS companies, agencies running client work through one API key, internal platform teams serving multiple product teams from a single model default.

How to package the offer

1. Spend routing audit

Pull recent usage logs, bucket requests by task type and complexity, and flag which ones ran on the expensive tier unnecessarily. Paid, fixed-scope, one to two weeks.

2. Routing layer build

Stand up a thin routing layer (LiteLLM, OpenRouter, or a Vercel AI Gateway config) that sends routine requests to the cheaper tier by default and escalates only when a task actually needs the flagship model's autonomy or context.

3. Escalation rules

Define the small number of conditions that should route to the expensive model — long-running autonomous work, tasks that already failed once, or high-stakes output — mirroring Anthropic's own "days-long, very autonomous projects" framing.

4. Quarterly re-tuning retainer

Every time a lab ships a new price tier, the optimal routing rules shift. This is the recurring revenue: revisit the split every quarter as new models launch and prices move.

Why the angle is stronger than generic "AI cost optimization"

Because you now have a concrete, citable, currently-happening example to open the conversation with: Anthropic's own flagship model got overtaken by its own cheaper model in corporate spending within a month of launch. If that is true inside Anthropic's own customer base, it is almost certainly true inside your prospect's AI bill too.

Bottom line

The AI market has quietly shifted from "which model is smartest" to "which model survives a routing decision." Companies that haven't built that routing logic on purpose are overpaying right now, and most of them don't know it.

Sell the audit that shows them the gap, then sell the layer that closes it.

Sources:

https://www.implicator.ai/anthropic-opus-5-overtakes-fable-5-corporate-spending/

https://ramp.com/data/ai-index-august-2026

https://news.ycombinator.com/item?id=49411102