Methodology & sources — Tokentrendr

Methodology

How Tokentrendr measures the market

Plain-English explanation of where our numbers come from and how the inversion is calculated.

Source of truth

Each provider's official pricing page. We scrape the structured data and store one snapshot per refresh.

Refresh cadence

A scheduled job runs every 6 hours, scrapes each pricing page, and inserts a new price_snapshots row.

History

Snapshots are append-only. Range selectors (24h / 7d / 30d / 90d) filter the slice used for charts and deltas.

The inversion rule

Most market dashboards color rising prices green. For buyers of AI tokens, that's backwards — rising token prices are bad for you. So we invert the convention:

Down = good

Green downward arrow. Token cost dropped versus the start of the selected range. You pay less.

Up = bad

Red upward arrow. Token cost rose versus the start of the selected range. You pay more.

Flat

Within ±0.05% of the baseline. Effectively unchanged for the range.

Formula

delta% = (latest_input_per_mtok − baseline_input_per_mtok) / baseline_input_per_mtok × 100

baseline = first snapshot inside the selected range. latest = most recent snapshot.

Sources

Where each provider's prices are obtained

Provider	Pricing page	Models tracked	Refresh
OpenAI	openai.com/api/pricing	3 models	every 6h
Anthropic	www.anthropic.com/pricing	3 models	every 6h
Google	ai.google.dev/pricing	3 models	every 6h
xAI	x.ai/api	2 models	every 6h
Mistral	mistral.ai/pricing	2 models	every 6h
Cohere	cohere.com/pricing	2 models	every 6h
DeepSeek	platform.deepseek.com/pricing	2 models	every 6h
Meta	llama.meta.com	2 models	every 6h
Perplexity	docs.perplexity.ai/guides/pricing	1 models	every 6h

Normalization

• All prices are converted to USD per 1,000,000 tokens.
• Per-1K listings are multiplied by 1,000; per-token by 1,000,000.
• Cached input and Batch API tiers are stored as separate fields where published.
• Missing fields are not guessed — we display "—".

Stack

• Extraction: Firecrawl structured JSON scrape against each pricing page.
• Storage: Lovable Cloud (managed Postgres), append-only price_snapshots.
• Schedule: pg_cron POSTs to /api/public/cron/refresh-prices every 6 hours.
• Deltas are computed at read time from the selected range.

Disclaimers

• Tokentrendr is an independent information site. We are not affiliated with any listed provider.
• Pricing pages can change between refreshes. Always confirm prices on the provider's site before contractual decisions.
• Recent news items are summarized from public sources linked on each card.