Transcription Models

Seven models behind one endpoint. The model parameter selects one, and every model returns the same output formats.

Choosing Your Model

All models are accessed via the same secure API infrastructure but cater to different priorities:

Current models

New integrations should start with Base or the Oracle family.

KolWrite Base

Low-cost transcription for clear audio, batch workloads, and cost-sensitive integrations.

Word-level timestamps Regular, Pro, and channel speaker separation Best for high-volume, straightforward audio

Oracle Mini

The fastest, lowest-cost Oracle tier for drafts and high-throughput transcription.

Fast Oracle processing Multilingual transcription and translation Optional diarization and channel separation with aligned timestamps

Oracle

The flagship Oracle tier for production transcription and multilingual workflows.

Production-quality transcription Translation to supported target languages Regular, Pro, and channel speaker separation

Oracle Pro

Premium Oracle quality for demanding production audio and stronger accuracy requirements.

Premium transcription quality Multilingual transcription and translation Regular, Pro, and channel speaker separation

Oracle Max

The highest-quality tier for difficult audio and work where transcription accuracy matters most.

Maximum transcription quality Designed for challenging audio Regular, Pro, and channel speaker separation

Legacy models

Maintained for compatibility with existing legacy integrations.

Core Legacy

A maintained legacy model for existing integrations that already depend on Core behavior.

Word-level timestamps Regular, Pro, and channel speaker separation Best for existing Core integrations

Edge Legacy

The previous-generation high-accuracy option, maintained for compatibility with existing integrations.

Word-level timestamps Regular, Pro, and channel speaker separation Best for existing Edge integrations

Model usage credits

Model Credits per hour
Current models
KolWrite Base
0.40
KolWrite Oracle Mini
0.80
KolWrite Oracle
1.20
KolWrite Oracle Pro
1.40
KolWrite Oracle Max
3.00
Legacy models
KolWrite Core
1.00
KolWrite Edge
1.20

Usage is based on audio duration and calculated per second with no minimum charge. 1 credit is the API billing unit previously displayed as $1.00 in the API console. Eligible Israeli billing accounts purchase at a net price of ILS 3.60 per credit before VAT; checkout shows VAT and the final ILS total. Other accounts purchase usage credits in USD at USD 1.20 per credit unless separate order terms apply.

Add-on usage credits

Add-on Credits per hour
Speaker Diarization
0.00
Pro Speaker Diarization
0.20
Standard Channel Separation
Included

Overlapping-speech separation is billed at the selected model rate for the greater of the recording duration or the total detected speech time across its physical channels. No diarization add-on applies. Oracle Mini automatically enables timestamp alignment when a speaker mode is selected.

Benchmark Performance

Accuracy and Word Error Rate (WER) on common benchmarks demonstrate the relative performance of our models. Higher accuracy is better; lower WER is better.

Accuracy (%)

Benchmark KolWrite Oracle KolWrite Edge KolWrite Core
Google FLEURS 86% 84.3% 82.2%

Word Error Rate (WER %)

Benchmark KolWrite Oracle KolWrite Edge KolWrite Core
Google FLEURS 14% 15.7% 17.8%

Note: Performance can vary based on audio quality, language, and specific domain.

All API interactions are secured using industry-standard protocols. KolWrite is committed to data privacy and does not use customer data submitted through the API to train our models.

Frequently Asked Questions

Find answers to common questions about our transcription API and models.