← Case studies
E-commerce

How Leaders in E-commerce Got Frontier-Model Quality at 1% of the Cost

Airbnb found that a fine-tuned classifier with 2 billion effective parameters matches frontier-level quality in under 100 ms at roughly 1% of the cost.

Airbnb fine-tuned Gemma4-E2B, a model with 2 billion effective parameters, on hundreds of thousands of domain-specific examples. Its max F1 rose from 0.563 to 0.902, on par with the best proprietary models, at a median latency of 35 ms against about 900 ms for GPT-6.1 Sol and Claude Sonnet 5.5 in low-thinking mode. At thousands of requests per minute, the frontier models using the concise prompt cost two orders of magnitude more than a single self-hosted GPU.

0.563 → 0.902
max F1 after fine-tuning Gemma4-E2B on hundreds of thousands of domain-specific examples
35 ms
median latency of the fine-tuned model
~900 ms
median latency of GPT-6.1 Sol and Claude Sonnet 5.5 in low-thinking mode
2 orders
of magnitude: what frontier models cost compared with a single self-hosted GPU, at thousands of requests per minute

The problem

Airbnb uses AI models for classification tasks. Frontier models are often good at these, but interactive products and high-volume workloads also need low latency, manageable serving costs, and control over the tradeoff between false positives and false negatives.

What they did

The task screens content for violations of Airbnb’s policy on pricing and payment transparency. The model is called thousands of times per minute, and a score must return within 200 ms.

The result

The fine-tuned classifier returned a score in about 85 ms p99 and is on par with the best proprietary models. Lower-cost proprietary options still run nearly five times the cost of the GPU, with lower accuracy than the fine-tuned model.

Caveats

Below roughly 15 to 30 requests per minute, Claude Sonnet 5.5 and GPT-6.1 Sol are cheaper than a dedicated GPU. Costs use public list prices as of October 2026.

Figures come from the source below. Anything from another place is attributed in the text.

Source: Airbnb Engineering: Frontier quality at 1% of the cost ↗
Ready to make AI more efficient?

Make AI
pay for itself.

Most clients uncover significant savings within the first two weeks. Let's start with yours — no commitment required.