How the charts are estimated

What we compare

Charts on the home, pricing, and environment pages use one shared model: a full exchange of about ~3,000 tokens (input + output), treated as one “query” or “exchange.”

Usage cost (pricing & home)

• Vivral Pro is shown at $4.99/month unlimited — the plan the charts use.
• Limited frontier plan is a typical consumer chat subscription of about $20/month. These plans are rate-capped / message-limited, not unlimited.
• Unlimited frontier path uses public API token prices: average of GPT-5-class and Claude flagship rates (~$5 / ~$27.50 per million input/output tokens). At a ~2:1 input:output mix that blends to about $12.50 per million tokens.
• Daily companion volume: about 40 exchanges/day × ~3,000 tokens ≈ 3.6 million tokens/month → about $45/month, or ~$540/year.
• That is roughly 9× Vivral Pro’s ~$60/year ($4.99 × 12).

Energy (environment)

• Frontier models are treated as ~3–4 trillion parameters class (interactive serving with data-center overhead).
• Cloud path energy is estimated at about 4.0 Wh (0.004 kWh) per exchange, scaled from published large-model energy measurements plus cooling (PUE) and network overhead.
• Vivral routine inference is modeled as on-device on iPhone GPUs (Neural Engine / GPU path), about 0.14 Wh per exchange under our mobile inference assumptions.
• That is about 96% less energy per query (~28× lower).

Electricity price

Where we convert on-device energy to dollars, we use average New Jersey residential electricity (about 23.5¢/kWh, EIA). At ~1,200 queries/month that is only about $0.04/month of electricity — negligible next to a $4.99 subscription — so usage-cost charts show the flat Pro price.

We do not publish internal model size on marketing charts; the point is architecture (remote frontier serving vs local inference), not a parameter count.