Breaking

Some Anthropic, OpenAI Customers Cut AI Costs Over 90%

June 24, 2026 at 11:30 EDT

Foundation Models
Sales & Business
Infra & Chips

Some large customers of Anthropic and OpenAI are cutting their AI bills by more than 90% by routing tasks from costly flagship models to cheaper ones offered by the same providers. The shift comes as cheaper models prove "good enough" for many tasks, prompting enterprises to mix and match models.

Enterprise AI · Cost Optimization

Companies Are Slashing AI Bills 90%+ by Routing to "Good-Enough" Cheaper Models

Major OpenAI and Anthropic customers are steering extraction, classification, and simple tasks away from premium flagships — one hospital software provider with a $100M AI budget used a model 23× cheaper and got satisfactory results.

90%+

Cost cut achieved by routing tasks to cheaper models

23×

Cheaper model used by a $100M-budget hospital software provider

50%

Extra Batch API discount, stackable with up to 90% prompt-cache savings

Output price per 1M tokens — flagship vs. budget tier

Same task, very different bill. Taller column = costlier.

$30

GPT-5.5
flagship

$25

Claude Opus 4.6
flagship

$15

Claude Sonnet 4.6
mid

Claude Haiku 4.5
budget

$4.50

GPT-5.4 mini
budget

Takeaway: A budget-tier model can cost roughly 6–7× less per output token than a flagship — before stacking caching and batch discounts.

The new default: route every task to the cheapest model that's good enough

Incoming task

→

Model router scores difficulty

→

Pick cheapest sufficient model

Cheap is "good enough"

Extraction, classification and simple processing are shifting to budget models — slashing bills as 2026 API costs surge and margins tighten.

Flagships still required

Complex reasoning and coding still need top models — making routing accuracy the key to balancing cost against quality.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.