Less searching. Less browsing = Less compute.
Compliant discovery, comparison and handoffs through MCP or HTTP. Less page parsing. Fewer tool loops.
One Relay call can replace many independent web interactions.
Keep your model, interface and final action.
See the economics of your workload.
Compare avoided inference with your total Relay fee and the work that remains.
An illustrative month with Opus 5.5
10,000 monthly active users × ten commerce tasks = 100,000 tasks.
Assume a 60% net reduction in the modeled input component, after retained and replacement input. This is an illustrative assumption, not a measured result.
- Original screenshot input
- $10,368.00
- Remaining or replacement input
- $4,147.20
- Modeled avoided input cost
- $6,220.80
- Growth monthly minimum
- $2,000.00
- Modeled benefit after the minimumBefore additional capacity and other changed costs
- $4,220.80
20 screenshots per task · 1000 × 1000 pixels · 1,296 tokens per image · $4 per million input tokens. Each image is billed once. Sources checked October 12, 2026: Anthropic pricing and image tokens.
Compare models and screenshot sizes
Same 100,000-task workload and 20 screenshots per task. These are input-price sensitivities, not evidence that each model achieves the same task quality.
| Model and rate condition | 1000 × 1000 | 1920 × 1080 |
|---|---|---|
| Claude Opus 5.5Standard uncached input | $10,368.00 | $21,528.00 |
| Claude Sonnet 5.5Standard uncached input | $5,184.00 | $10,764.00 |
| GPT-5.6 SolStandard uncached input, up to 272K prompt tokens | $9,832.00 | $19,584.00 |
| GPT-5.6 TerraStandard uncached input, up to 272K prompt tokens | $4,916.00 | $9,792.00 |
Provider-specific image token calculations: Claude uses 1,296 / 2,691 tokens (high resolution at 1920 × 1080); OpenAI uses 1,229 / 2,448 tokens at original detail. Standard input rates, with each screenshot billed once. Sources: Claude pricing, Claude vision, OpenAI pricing and OpenAI vision.
Integration and pricing.
Plans grow with your monthly active users and include agreed API usage.