Last verified
PERCEPTION MODELIMAGE + VIDEO + AUDIOGROUNDED OUTPUT36K CONTEXTCACHE 75% OFF

Perceptron Mk1.5 API Pricing

Perceptron Mk1.5 is a perception model for visual and physical agents: it reads images, video and audio and answers with text plus grounded annotations - points, boxes, polygons, time clips and object tracks - or with function calls. Perceptron prices it at $0.15/M input, $1.50/M output and $0.0375/M cached input, the same card as Mk1. Launched September 25, 2026. Checked by hand against docs.perceptron.inc.

Input - per 1M tokens
$0.15/M
Media billed as tokens = Mk1
Output - per 1M tokens
$1.50/M
Max output 8,192 = Mk1
Cached input - 75% off
$0.0375/M
Multilook shared context -75%
Effective - agentic blend
$0.173/M
92/8 split - 82% cache
§ 01 / TERMINAL

Run the numbers.

Live calculator pre-loaded with Mk1.5's rates. Images, video frames and audio are all billed as input tokens - about 750 tokens per minute of audio - so size the input side from Perceptron's tokenization guide, not from word counts. The cached rate applies to context reused across Multilook prompts.

$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
Open full calculator (all models · share URL · CSV) →
§ 02 / SCENARIOS

Real-world presets.

DETECTION

Shelf image object boxes

$0.00102/image
2,000 in - 600 out~98,473 images/$100
VIDEO

Clip event timing

$0.00341/clip
20,000 in - 1,500 out~29,368 clips/$100
AUDIO

10-minute call transcript

$0.00343/call
7,500 in - 2,000 out~29,127 calls/$100
OCR

Receipt field extraction

$0.00069/doc
1,500 in - 400 out~145,639 docs/$100
§ 03 / TOKENIZER

Paste text. See tokens. See cost.

What's a token →
Estimate · perceptron-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type

This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.

Characters 691
Words 112
Tokens (estimated) 179 tokens
Cost as input · uncached $0.00003 USD
Cost as output · uncached $0.00027 USD
Cost as cached input $0.00001 USD
§ 04 / SHELF

Up against the shelf.

All models →
Model Input /M Output /M Effective blended Context Best for
Perceptron Mk1.5 Current $0.15 cache $0.0375 $1.50 $0.173 agentic 92/8 36K Grounded image, video and audio perception
Perceptron Mk1 $0.15 cache $0.0375 $1.50 $0.173 same 32K Earlier Mk1, same price, no audio
Gemini 3.5 Flash-Lite $0.30 cache $0.03 $2.50 $0.272 +57% 1M Budget general multimodal model
Doubao Seed 2.1 Lite $0.119 cache $0.0238 $0.402 $0.0698 −2.5× 1M Low-cost omni-modal model
Step 3.7 Flash $0.20 cache $0.04 $1.15 $0.155 −10% 256K Video-capable reasoning model
Qwen3.8-Omni-Flash $0.15 cache $0.016 $0.47 $0.0745 −2.3× 1M Omni-modal model with audio

Frequently asked.

Practical Perceptron Mk1.5 pricing questions - what a token is when the input is video or audio, and what the cache applies to.

Q · 01 What is Perceptron Mk1.5 priced at? +
Perceptron's model card lists perceptron-mk1.5 at $0.15/M input, $1.50/M output and $0.0375/M cached input - the same three rates as Mk1.
Q · 02 How are images, video and audio billed? +
As input tokens at the input rate. Audio runs about 750 tokens per minute, capped at 16,384 audio tokens (about 21.8 minutes) per item; images and video frames are tokenized by size and frame count per Perceptron's tokenization guide. Everything - media, prompt and the answer - shares one 36,864-token context.
Q · 03 When does the cached rate apply? +
To prompt tokens the API reports as cached, which in practice means Multilook: one request sends a shared context plus up to 16 independent prompts, the context is prefilled once, and reused tokens bill at $0.0375/M. Perceptron notes cached tokens are a subset of prompt tokens, so do not add them to the prompt total twice.
Q · 04 What does it return besides text? +
Grounded annotations - points, boxes and polygons on images, time clips on video and audio, and timestamped object tracks - plus JSON Schema or regex-constrained answers and function calls. Function tools cannot be combined with constrained output in the same request.
Q · 05 How is it different from Mk1? +
Same price. Mk1.5 adds audio input (and video soundtracks), function calling, video tracking and a reasoning_effort control, and its context is 36,864 tokens against Mk1's 32K. Mk1 remains available.
Q · 06 Can I run it myself? +
No. Perceptron describes its Mk models as closed-source, sold through its own API and resold by third parties. The earlier open Isaac 0.1 and 0.2 models were retired from Perceptron's API on September 25, 2026.
Y
Reviewed by Yaroslav Vikhariev Founder · AI//COST · Prices checked by hand against docs.perceptron.inc · Last verified Sep 29, 2026
Methodology Report a correction More by Y.V.