OpenAI's Whisper kicked off the speech-to-text revolution in 2022. In 2026, the market has matured: a dozen providers offer Whisper-based APIs with different pricing, speed, and accuracy tradeoffs. The cheapest option isn't always the best β language support, latency, and accuracy vary widely.
This guide covers the major Whisper-based APIs, their current pricing, and which to pick for your specific use case.
| Provider | Price/min | Speed | Languages | Best for |
|---|---|---|---|---|
| OpenAI Whisper API | $0.006 | Medium | 99 | Reliability |
| Groq Whisper | $0.0033 | 10x faster | English-heavy | Real-time |
| AssemblyAI Universal | $0.015 | Fast | 99 | Accuracy + features |
| Deepgram Nova | $0.0043 | Fast | 36 | Streaming |
| Google Cloud Speech | $0.016 | Fast | 125+ | Google ecosystem |
| Azure Speech | $0.016 | Fast | 100+ | Enterprise |
| Alibaba Cloud ASR | $0.0014 | Fast | Chinese-strong | Chinese content |
Prices are USD per minute of audio. Free tiers exist for OpenAI (10 min), AssemblyAI (100 min), and Alibaba (60 min).
Price: $0.006/min ($0.36/hour of audio)
Model: whisper-1 (the only model offered)
Speed: Medium (a 1-hour file takes ~3-4 minutes to transcribe)
Languages: 99, with auto-detect
Free tier: None for new accounts (was $5 trial in 2023)
Output: JSON, SRT, VTT, plain text
Pros: The reference implementation. Most reliable. Outputs SRT and VTT directly, which is what most video tools want.
Cons: Slowest of the paid options. No streaming support. No speaker diarization (who said what).
Price: $0.0033/min (~$0.20/hour)
Model: whisper-large-v3 on Groq's LPU hardware
Speed: 10x faster than OpenAI (1-hour file in 20 seconds)
Languages: Best on English, good on major languages
Free tier: Yes, generous
Pros: Fastest API. Same accuracy as OpenAI (uses same model). Cheapest option with free tier.
Cons: Limited language support compared to OpenAI. No native SRT output (need to format timestamps yourself).
Price: $0.015/min ($0.90/hour) for Universal, $0.006/min for
base
Speed: Fast (similar to OpenAI)
Languages: 99
Features: Speaker diarization, sentiment analysis, entity detection,
auto-chapters, profanity filtering
Pros: Most feature-rich API. Built-in speaker labels ("Speaker A: hello", "Speaker B: hi"). Excellent accuracy on noisy audio.
Cons: More expensive. Overkill if you just need captions.
Price: $0.0043/min (~$0.26/hour)
Speed: Real-time streaming available
Languages: 36 (English best)
Features: Streaming, speaker diarization, custom models
Pros: Real-time streaming. Custom vocabulary. Good for live captioning where you can't wait for batch processing.
Cons: Smaller language list. Best for English-only projects.
Price: $0.0014/min for short audio, $0.0021/min for long audio
(Chinese yuan conversion: Β₯0.0001/second)
Languages: Chinese (best), English, Japanese, Korean, others
Free tier: 60 minutes/month
Pros: Cheapest option for Chinese content. Best Chinese accuracy (handles dialects, technical jargon). Domestic Chinese cloud = no API access issues.
Cons: Weaker on English and other languages. Account requires Chinese real-name verification for production use.
Use Groq free tier. It's fast, free, accurate enough, and handles English perfectly. Format the response to SRT yourself (it's just JSON with timestamps).
Use OpenAI Whisper API. Pay for reliability. Native SRT output saves you formatting work. Document your fallback if OpenAI has an outage.
Use Alibaba Cloud. Accuracy is best in class for Mandarin and Chinese dialects, and the price is half of OpenAI.
Use Deepgram streaming or run LocalVocal locally with Whisper. Cloud APIs add too much latency for live use.
Use AssemblyAI. Best speaker diarization and language detection. Worth the premium if you're doing interview/podcast transcription.
Quick math for common use cases:
| Use case | Volume | OpenAI | Groq | AssemblyAI |
|---|---|---|---|---|
| Weekly YouTube video (10 min) | 40 min/month | $0.24 | $0.13 | $0.60 |
| Daily podcast (30 min) | 15 hours/month | $5.40 | $2.97 | $13.50 |
| Online course (100 videos, 5 min each) | 500 min total | $3.00 | $1.65 | $7.50 |
| Transcription business (1000 hours/year) | 1000 hours | $360 | $198 | $900 |
If you're processing more than 500 hours/month of audio, self-hosting Whisper becomes cheaper:
Self-hosting only makes sense if you have a developer to set it up and maintain it. For one-off jobs or small businesses, the API is the better deal.
No β Groq is cheaper and faster in 2026. OpenAI's advantage is ecosystem (integration with GPT-4o for cleanup) and brand trust.
No, the free trial ended in 2024. You need a paid OpenAI account with at least $5 credit.
AssemblyAI Universal handles noisy audio best. It uses additional models for denoising and accent detection that the base Whisper API doesn't have.
Groq's free tier is generous (enough for personal use). For unlimited free, run Whisper locally β there's no API quota.
If you just need a one-off transcription, use Groq's free tier. If you need production reliability, use OpenAI's Whisper API. If you need Chinese content, use Alibaba Cloud.