Deepgram Alternatives

AssemblyAI Universal-3.5 Pro cuts Deepgram Nova-3’s $0.258 per audio hour to $0.21 and AA-WER from 5.2% to 3.0%, and ElevenLabs Scribe v2 reaches 2.2% at $0.22; none of the alternatives matches Nova-3’s 548.1× speed factor.

Updated · 7 sources · Methodology

NeedPickWhy
Lowest example costAssemblyAI$0.21 / h3
Fastest measured outputDeepgram548.1×1
AA-WER v2ElevenLabs2.2%1

Performance

ProviderModelAA-WER v2 (%)Speed factor
Deepgram1Nova-35.2%548.1
AssemblyAI1Universal-3.5 Pro3.0%100.3
ElevenLabs1Scribe v22.2%68.4
OpenAI1GPT Transcribe3.3%52.2

Prices and request cost

ProviderPrice / audio hour100 h
Deepgram2$0.258$25.80
AssemblyAI3$0.21$21.00
ElevenLabs4$0.22$22.00
OpenAI5$0.27$27.00
Cost example: 100 hours of mono pre-recorded audio
Deepgram$25.80
AssemblyAI$21.00
ElevenLabs$22.00
OpenAI$27.00

Options, strengths and limitations

  1. AssemblyAI · Universal-3.5 Pro

    Universal-3.5 Pro lowers 100 audio hours from Deepgram’s $25.80 to $21.00 and AA-WER from 5.2% to 3.0%, processing at 100.3× real time.31

    Price / audio hour
    $0.21
    AA-WER v2
    3.0%
    Speed factor
    100.3×

    Strengths

    Lower hourly price and lower AA-WER than Nova-3 in this comparison.13

    Limitations

    Universal-3.5 Pro covers 18 languages; other languages can use a different model, so this row’s accuracy does not extend to all AssemblyAI languages.7

  2. ElevenLabs · Scribe v2

    Scribe v2 has the lowest AA-WER here at 2.2%, less than half of Nova-3’s 5.2%, for $22.00 per 100 audio hours.41

    Price / audio hour
    $0.22
    AA-WER v2
    2.2%
    Speed factor
    68.4×

    Strengths

    The lowest word error rate among the selected transcription models.1

    Limitations

    Scribe v2 processes at 68.4× in AA, below Nova-3’s 548.1×; its accuracy lead comes with slower batch processing here.1

  3. OpenAI · GPT Transcribe

    GPT Transcribe costs $27.00 per 100 audio hours, $1.20 more than Deepgram, at 3.3% AA-WER and 52.2× real time.51

    Price / audio hour
    $0.27
    AA-WER v2
    3.3%
    Speed factor
    52.2×

    Strengths

    Audio transcription is available through the same provider account as GPT APIs.6

    Limitations

    GPT Transcribe accepts files up to 25 MB, so long recordings must be compressed or split, and its 52.2× speed factor is the lowest here. It replaces GPT-4o Transcribe, which OpenAI shuts down on 2027-02-26.6

Methodology and sources

AA-WER v2 uses AA-AgentTalk (50%), cleaned VoxPopuli (25%) and cleaned Earnings22 (25%). Speed factor is AA’s 7-day median of audio seconds processed per second on 10-minute audio, not realtime latency. Prices are base mono pre-recorded rates; add-ons, extra channels and realtime models are billed separately.

Audio charges are estimates, not measured invoices; tax and subscription credits are excluded. Novita AI publishes this speech API comparison using public sources.

Alternative cards are ordered by base price per audio hour: AssemblyAI, ElevenLabs, then OpenAI. Accuracy and batch speed are separate picks. OpenAI bills GPT Transcribe by audio duration at $0.0045 per minute, which is $0.27 per hour.3456

  1. Artificial Analysis · AA-WER v2 non-streaming
  2. Deepgram · official API pricing
  3. AssemblyAI · official API pricing
  4. ElevenLabs · official API pricing
  5. OpenAI · official API pricing
  6. OpenAI · file transcription and model retirement
  7. AssemblyAI · Pro language scope

Build with Novita Model APIs

Novita AI doesn’t offer speech-to-text. Pair your transcription provider with Novita LLM APIs to summarize and analyze transcripts, or Fish Audio text-to-speech for voice output.

Start BuildingExplore Models