How we make this list: we compare pricing pages, product documentation and public user reviews. Ratings come from G2, Capterra or Trustpilot, with the date we checked them.
Key takeaways
- Deepgram's Nova-3 batch pricing near $0.0043 a minute and streaming near $0.0077 a minute set a low bar for real-time voice agents.
- AssemblyAI is the cheapest batch option at $0.0025 a minute on Universal-2, with a $50 free credit and strong diarization and language coverage.
- OpenAI Whisper API is the simplest choice at $0.006 a minute, though it lacks real-time streaming and caps files at 25 MB.
- Google Cloud, Azure, and AWS suit teams already on those clouds, with batch discounts that bring costs well below list rates at scale.
- Rev AI, Speechmatics, and Gladia cover human-assisted accuracy, multilingual batch work, and predictable unit economics for global teams.
Quick comparison
Is your product missing or out of date?
We review this list regularly. Vendors can request a listing, send updated pricing or ask about featured placement.
The best Deepgram alternatives
#1AssemblyAI
Best for: Cheap batch plus voice agents4.3/5 on
G2 (38 reviews, checked Oct 2026)
Visit assemblyai.com
Starting price$0.0025 per minute
Free planYes
AssemblyAI is the strongest all-round Deepgram alternative. Its Universal-2 batch model costs $0.0025 a minute, or $0.15 an hour, undercutting Deepgram's Nova-3 batch rate of $0.0043 a minute, and a $50 free credit covers roughly 333 hours of batch audio. Universal-3.5 Pro and the Universal-3.6 Pro Realtime model add accuracy for voice agents, with streaming near $0.45 an hour on the Pro realtime model and no annual commitment.
The platform also bundles speech understanding: diarization, keyterms prompting, real-time prompting, and unlimited concurrency suit production voice AI apps. Add-ons stack though, with PII redaction and entity detection each carrying their own per-hour fee, so a cheap API call with three add-ons can double. For a developer team building English-first voice agents or batch transcription, AssemblyAI is the price and feature leader.
What we like- Lowest batch rate at $0.0025 a minute on Universal-2
- $50 free credit
- 99 languages
- Realtime streaming with diarization and keyterms prompting
Watch out for- Add-ons such as PII redaction and entity detection carry separate per-hour fees
- Streaming rates run higher than batch
Starting price$0.006 per minute
Free planNo
OpenAI's Whisper API is the simplest path to transcription. The whisper-1 model costs $0.006 a minute across 99 languages, and gpt-4o-mini-transcribe brings that near $0.003 a minute with token billing. There is no model to host, no infrastructure to manage, and integration takes an afternoon for most teams.
The limits are real: whisper-1 has no real-time streaming mode, and the 25 MB file limit forces chunking on long recordings. Teams needing live transcription should use the Realtime API instead, where gpt-4o-transcribe runs near $0.006 a minute and the mini variant near $0.003. For a team already built around OpenAI APIs that needs batch transcription without fuss, Whisper is the obvious pick.
What we like- Simple API with 99 languages
- Gpt-4o-mini-transcribe near $0.003 a minute
- No model management needed
Watch out for- No real-time streaming on whisper-1
- 25 MB file limit forces chunking on long recordings
- No free tier
Starting price$0.003 per minute
Free planNo
Rev AI pairs a competitive AI model with a human option. Its Reverb ASR model runs near $0.003 a minute for async work, with foreign-language rates near $0.005, and 58 plus languages async. For legal, medical, or broadcast work where accuracy is non-negotiable, human review is available, though human transcription costs orders of magnitude more than AI-only.
Streaming covers only 9 languages, so real-time global work belongs elsewhere. The free offer is a 5-hour trial rather than a standing free plan. For a team that needs AI speed with a human safety net for critical files, Rev AI is the hybrid choice.
What we like- Reverb ASR near $0.003 a minute
- Human review option for maximum accuracy
- 58 plus languages async
Watch out for- Only a 5-hour free trial, no standing free plan
- Streaming covers 9 languages
- Human transcription costs far more than AI-only
Starting price$0.016 per minute
Free planYes
Google Cloud Speech-to-Text suits teams already on Google Cloud. Its V2 standard recognition starts near $0.016 a minute with 125 plus languages and the Chirp 2 model, while dynamic batch drops near $0.003 a minute at volume, which changes the math for large async queues. A 60-minute monthly free tier covers light testing.
Outside Google Cloud, the setup overhead and standard rate make it less attractive than AssemblyAI or Deepgram. Inside GCP, though, the integration, IAM, and data residency story is strong. For a GCP-native team transcribing large async queues, Google's dynamic batch pricing is competitive.
What we like- 125 plus languages
- Chirp 2 model
- Dynamic batch near $0.003 a minute at volume
- Deep GCP integration
Watch out for- Standard V2 rate near $0.016 a minute is high for small volumes
- GCP overhead adds complexity outside Google Cloud
Starting price$0.003 per minute batch
Free planYes
Azure AI Speech covers the full range: real-time, fast, and batch transcription across 100 plus languages, with container deployment for teams that need on-premises or edge processing. Batch rates near $0.003 a minute make it one of the cheapest batch options, and a 5-hour monthly free tier on the F0 tier covers evaluation.
Real-time rates near $0.017 a minute are higher, and pricing varies by region and agreement, so enterprise buyers should confirm figures. Custom speech models let teams tune accuracy for accents and domain vocabulary. For a Microsoft estate or a team needing container deployment, Azure is the natural Deepgram alternative.
What we like- Batch transcription near $0.003 a minute
- 100 plus languages
- Container deployment options
- Custom speech models
Watch out for- Real-time rates near $0.017 a minute
- Pricing varies by region and agreement
- Microsoft-centric setup
Starting price$0.024 per minute
Free planYes
AWS Transcribe is the default for AWS infrastructure. It integrates with S3, Lambda, and the wider AWS stack, supports medical vocabulary for healthcare workloads, and offers both streaming and batch modes. A 60-minute monthly free tier runs for 12 months, and scale discounts bring the standard rate from $0.024 a minute down toward $0.0078 at volume.
The standard rate is the highest on this list, and language coverage at 31 plus is narrower than Google's or Azure's. For a team already on AWS with audio landing in S3, Transcribe's integration outweighs its list price. For a standalone transcription need, AssemblyAI or Deepgram costs less.
What we like- Deep AWS integration with S3 and Lambda
- Medical vocabulary support
- Streaming and batch
- Scale discounts to $0.0078 a minute
Watch out for- Standard rate near $0.024 a minute is the highest here
- 31 plus languages is narrower
- Setup is AWS-heavy
Starting price$0.004 per minute
Free planYes
Speechmatics offers a clean batch story. Its Melia-1 model from 2026 runs near $0.004 a minute, or $0.24 an hour, across 50 plus languages, with 480 minutes a month free. Custom dictionaries improve accuracy on names and domain terms, and translation support helps global teams.
Melia-1 is batch only, so real-time work needs the Ursa-2 model near $0.017 a minute or the Enhanced model near $0.0125 a minute, both with the same 480-minute free tier. For a team transcribing recorded content in many languages on a predictable budget, Speechmatics is a strong mid-price option.
What we like- Melia-1 batch model near $0.004 a minute
- 480 minutes a month free
- 50 plus languages
- Custom dictionary and translation
Watch out for- Melia-1 is batch only
- Real-time Ursa-2 runs near $0.017 a minute
- Enterprise pricing is custom
Starting price$0.0102 per minute
Free planYes
Gladia targets global teams that need predictable unit economics. Its Solaria-1 model covers 100 languages with streaming latency under 103 milliseconds, and a 10-hour monthly free tier covers evaluation. The standard rate near $0.0102 a minute is higher than AssemblyAI's batch rate, but the Growth tier near $0.0033 a minute brings it in line for committed volumes.
The pitch is cost modeling without add-on uncertainty: diarization and audio intelligence features are priced transparently, so a production bill at 1,000 hours a month stays predictable. For a team serving users across languages and geographies that wants to model costs before scaling, Gladia is worth shortlisting.
What we like- 100 languages
- Sub-103 millisecond streaming latency
- Growth tier near $0.0033 a minute
- Predictable costs without add-on uncertainty
Watch out for- Standard rate near $0.0102 a minute is higher than AssemblyAI batch
- Smaller ecosystem than the cloud giants
How we chose
We compared each tool against what Deepgram actually sells: fast, accurate speech-to-text APIs for real-time voice agents and high-volume batch work. Pricing figures come from vendor pages and published comparisons checked in October 2026, and anything we could not confirm is written as "See website". We favored actively maintained APIs that serve US developers, and we flagged where a tool is batch-only, streaming-only, or cloud-bound, since that decides whether it fits a given architecture.
What to check before you pick
Start with the mode you need: real-time voice agents, batch transcription, or both. For real-time work, shortlist AssemblyAI, Deepgram, and Gladia, and compare latency and streaming rates. For batch work, shortlist AssemblyAI, Azure batch, and Speechmatics, and compare per-minute rates plus add-on fees. Then run a benchmark on 20 to 30 files representative of your actual audio, since aggregate accuracy figures from lab conditions often miss accent, noise, and domain vocabulary effects. Check file-size limits, language coverage for your users, diarization costs, and data residency before committing, and confirm pricing on the vendor page, since speech-to-text rates moved several times during 2026.
Frequently asked questions
Is Deepgram still the cheapest speech-to-text API?
Not for batch work. AssemblyAI Universal-2 batch at $0.0025 a minute undercuts Deepgram Nova-3 batch near $0.0043 a minute. For streaming voice agents, Deepgram remains competitive, though AssemblyAI's realtime rates are close.
Which Deepgram alternative is best for voice agents?
AssemblyAI's Universal-3.6 Pro Realtime and Deepgram's Flux are the leading voice-agent models. AssemblyAI reports lower word error rates on its own benchmark, while Deepgram reports lower latency near 250 milliseconds. Run your own benchmark before choosing.
Can I self-host instead of using an API?
Yes. Open-source Whisper via WhisperX costs only compute, roughly $0.001 to $0.003 a minute self-hosted, but you manage GPUs, scaling, and model updates. Managed APIs trade that work for per-minute billing.
Do these APIs support diarization?
Most do. AssemblyAI, Deepgram, Rev AI, and Gladia offer speaker diarization, though some charge it as an add-on near $0.002 a minute. Confirm whether diarization is included in the base rate before budgeting.
What is the file-size limit for transcription APIs?
It varies: OpenAI Whisper caps files at 25 MB, Deepgram allows 2 GB or 10 hours, AssemblyAI allows 5 GB or 10 hours, and Gladia allows 5 GB or 8 hours. Long recordings may need chunking on stricter APIs.
How much does 100 hours of transcription cost?
At batch rates, AssemblyAI Universal-2 costs about $15, Deepgram Nova-3 batch about $26, and OpenAI whisper-1 about $36. Add-ons such as diarization and PII redaction can raise those figures, so confirm the loaded cost.
Sources we checked (8)
- https://www.assemblyai.com/pricing
- https://platform.openai.com/docs/pricing
- https://www.rev.com/pricing
- https://cloud.google.com/speech-to-text/pricing
- https://azure.microsoft.com/pricing/details/ai-speech/
- https://aws.amazon.com/transcribe/pricing/
- https://www.speechmatics.com/pricing
- https://www.gladia.io/pricing
Is your product missing or out of date?
We review this list regularly. Vendors can request a listing, send updated pricing or ask about featured placement.
Shihab Mia · Founder, Rankite
Shihab has 5+ years in B2B and SaaS SEO. He writes Rankite's software buying guides from pricing pages, documentation and public user reviews. More about Shihab.
OpenAI Whisper API
Rev AI
Google Cloud Speech-to-Text
Azure AI Speech
AWS Transcribe
Speechmatics
Gladia