Portfolio 12 · modality ledger

Multimodal AI API cost estimator

Apply provider-specific media preprocessing, accept official pasted counts where required, and keep every text, image, audio, video, document, cache, and output charge separate.

Decision this answers: Which provider bill is lower for the same counted workload after its native resizing and counting rules—not which model has equivalent quality?

Scenario

Decision summary

Uncached input

$14.55exact-derived

Text, image, audio, video, and document lines.

[1]

Cached input

$0.30exact-derived

Cached meter kept separate.

[1]

Output

$12.50exact-derived

Output and thinking-token meter.

[1]

Mixed-media bill

$27.35exact-derived

All visible modality line items reconciled.

[1][2][3]

Direct-labeled result profile

Uncached input Cached input Output Mixed-media bill

The tables below are the accessible source of truth; bar lengths never carry identity alone.

Provider counting comparison

Provider counting comparison
Candidate	Original media	Billable media	Counting authority	Unknown policy	Comparison claim	Evidence
Gemini	Visible	Provider-counted tokens	CountTokens/pricing	Paste count	Bill only	measured[1]
OpenAI	Visible	Resized/tiled dimensions	Vision guide/API count	Do not price	Bill only	measured[2]
Anthropic	Visible	Documented image count	Vision/token counting	Do not price	Bill only	measured[3]

Official surface versus this site

Gemini API pricing

Use the official surface to confirm native prices, counting rules, limits, and product eligibility.

This offline workbench

Official pages define model and modality meters. This page reconciles a complete mixed-media batch in one local ledger while preserving provider preprocessing and unknown counting rules.

Sources and methodology

Every result-affecting reference is visible here without JavaScript and is retained in the JSON export.

[1] Gemini API pricing

Authority: Google
Native identifier: gemini-3.5-flash-lite standard
Region: global
Unit: USD/1M tokens
Evidence: published list
Price kind: list
As of: 2026-08-01
Effective from: 2026-08-01
Retrieved: 2026-08-01
Freshness: fresh
Normalization: The dated $0.30 input, $0.03 cached input, and $2.50 output meters are applied to separate counted modalities.

Open the authoritative source

[2] OpenAI vision guide

Authority: OpenAI
Native identifier: vision input counting
Unit: resized pixels/tiles
Evidence: measured
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: Original and billable dimensions remain visible; unverified local counts are labeled estimates.

Open the authoritative source

[3] Anthropic vision guide

Authority: Anthropic
Native identifier: vision token counting
Unit: image tokens
Evidence: measured
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: Provider-specific media preprocessing is kept distinct and unknown rules cannot yield zero cost.

Open the authoritative source

Limits and non-claims

Cross-provider rows compare bills, not capability, quality, latency, or safety.
Model-dependent counts should come from the provider counting API and can be pasted locally.
Unsupported modalities remain visible and unpriced rather than inheriting a zero meter.