AI Video Token & Cost Calculator

A multimodal model doesn't "watch" your video — it converts it into tokens, the same billable units it uses for text, and charges them at its input-token rate. There are two very different regimes, and the gap between them can be 30×, so which model you pick matters far more than the length of the clip.

Gemini (native video). Gemini ingests the video file directly, sampling it at 1 frame per second and billing each frame as an image at a fixed 258 tokens (or 66 at low media resolution). The audio track, when included, is a separate 32 tokens per second:

• video_tokens = duration_seconds × 258
• audio_tokens = includeAudio ? duration_seconds × 32 : 0
• total = video_tokens + audio_tokens

OpenAI & Anthropic (frame-sampling). GPT-4o and Claude have no native video input, so you decode the video into still frames at a chosen frame rate and send each as an image:

• sampled_frames = ceil(duration_seconds × fps)
• GPT-4o per frame = 85 + 170 × tiles (512px tiles, detail=high)
• Claude per frame = round(width × height / 750), edge ≤ 1568px
• tokens = sampled_frames × per_frame_tokens

Audio is intentionally left out of the GPT-4o and Claude paths: their APIs accept audio through separate audio-input endpoints, so folding it into the frame count would double-charge. That keeps the visual-token comparison honest.

Cost. Each token stream is priced at its own published input rate — cost = tokens / 1,000,000 × rate — with Gemini's audio surcharge applied to the audio tokens only. The LKR figure multiplies by your exchange rate, and the monthly total multiplies by videos per month. A fits in contextcheck compares total tokens against each model's input window (about 1M for Gemini, 128k for GPT-4o, 200k for Claude). Every constant is cross-checked against the provider docs on each build, and a per-second formula independently verifies the Gemini frame-count path.

• Google — Gemini: understand and count tokens (video 1 fps, audio 32 tokens/s)
• Google — Gemini video understanding (native 1 fps sampling, media resolution)
• Google — Gemini API pricing (input rates, audio surcharge)
• OpenAI — Images and vision guide (frames billed as images, tiling)
• OpenAI — API pricing (GPT-4o input rate)
• Anthropic — Vision docs (width × height ÷ 750 per frame)
• Anthropic — Pricing (Claude Sonnet input rate)

Token rules and per-1M input rates were last cross-checked against these sources on 2026-07-05. Providers revise media pricing periodically — confirm against your latest invoice for exact billing.