Introducing WhisperOpenAIReleased March 1, 2023

Whisper

Universal speech recognition model trained on 680,000 hours of multilingual and multitask supervised audio data.

audio-transcriptionOpen Weights & API$0.006 / min audioContext: 30K (30,000 tokens)

Technical Specifications

Architecture Type
Sequence-to-Sequence Audio Transformer
Total Parameters
1.5B (large-v3 equivalent)
Context Window
30K (30,000 tokens)
Max Output Tokens
4.096K (4,096 tokens)
Knowledge Cutoff
May 2022
Supported Modalities
audio, speech, text
License & Access
MIT License (Weights) / OpenAI API Terms
Weights Formats
safetensors, onnx, gguf

Benchmark Evaluations

English Wer
4.2
Multilingual Average Wer
7.8

Deep Architectural Overview

Whisper revolutionized automated speech recognition (ASR). Robust against background noise, accents, and technical jargon, it provides multilingual transcription and English translation available both via API and open-source MIT weights.

Strengths & Considerations

Core Strengths
  • Exceptional noise and accent robustness
  • Freely available MIT open weights
  • Low API cost ($0.006/minute)
Known Limitations
  • Audio chunking required for files exceeding 25MB via API

Token & API Pricing

Input Tokens (1M)Free (Open Weights)
Output Tokens (1M)Free (Open Weights)
Self-Hosted Min VRAM4 GB
Pricing is verified directly against OpenAI's developer documentation and API rate sheets.

Similar & Alternative Models

Explore other frontier models from OpenAI and comparable reasoning engines.

Browse all models
OpenAI

OpenAI GPT-6 model for focused, high-volume workloads where cost efficiency is the priority.

1.05M ctxUSD 0.10 / 1M input tok; USD 0.50 / 1M output tok
OpenAI

OpenAI GPT-6 model built for complex coding and agentic workflows, balancing intelligence and cost.

1.05M ctxUSD 2 / 1M input tok; USD 10 / 1M output tok
OpenAI

OpenAI’s premier sixth-generation frontier model, built for the most demanding end-to-end software engineering and scientific research tasks.

1.05M ctx$10.00 / 1M tok (GPT-6 Frontier)