Avencera Get notified

Choose a model

Pick a model for the audio you have.

Different recordings need different models. Find yours below, with its price an hour. With or without speaker labels, every model costs less an hour than Microsoft Azure, ElevenLabs, AssemblyAI, OpenAI, Gladia and Speechmatics.

Kinds of audio

On every chart, top left is best: fewer errors and cheaper. The dashed line marks the fewest errors each price buys.

  • General audio

    A mix of recordings, or audio that fits none of the kinds below

    Qwen3-ASR $0.09 / hour$0.10 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 6.18% words errors,16.9% speakers errors,11.5% average errors,$0.213 an hour
    2. AvenceraQwen3-ASR, recommended 5.40% words errors,23.5% speakers errors,14.5% average errors,$0.10 an hour
    3. AvenceraGranite Speech 5.68% words errors,23.5% speakers errors,14.6% average errors,$0.09 an hour
    4. AvenceraParakeet v2 6.02% words errors,23.5% speakers errors,14.8% average errors,$0.07 an hour
    5. AvenceraParakeet v3 6.18% words errors,23.5% speakers errors,14.9% average errors,$0.08 an hour
    6. AssemblyAI 5.39% words errors,33.2% speakers errors,19.3% average errors,$0.23 an hour
    7. ElevenLabs 5.24% words errors,39.8% speakers errors,22.5% average errors,$0.22 an hour

  • Meetings

    Team calls, board meetings and stand-ups recorded in a room

    Granite Speech $0.08 / hour$0.09 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 9.42% words errors,16.9% speakers errors,13.2% average errors,$0.213 an hour
    2. AvenceraGranite Speech, recommended 6.96% words errors,23.5% speakers errors,15.2% average errors,$0.09 an hour
    3. AvenceraQwen3-ASR 8.31% words errors,23.5% speakers errors,15.9% average errors,$0.10 an hour
    4. AvenceraParakeet v2 9.09% words errors,23.5% speakers errors,16.3% average errors,$0.07 an hour
    5. AvenceraParakeet v3 9.42% words errors,23.5% speakers errors,16.5% average errors,$0.08 an hour
    6. AssemblyAI 10.63% words errors,33.2% speakers errors,21.9% average errors,$0.23 an hour
    7. ElevenLabs 8.19% words errors,39.8% speakers errors,24.0% average errors,$0.22 an hour

  • Calls and interviews

    Sales and support calls, interviews and user research

    Qwen3-ASR $0.09 / hour$0.10 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 13.44% words errors,16.9% speakers errors,15.2% average errors,$0.213 an hour
    2. AvenceraQwen3-ASR, recommended 12.14% words errors,23.5% speakers errors,17.8% average errors,$0.10 an hour
    3. AvenceraParakeet v3 13.44% words errors,23.5% speakers errors,18.5% average errors,$0.08 an hour
    4. AvenceraParakeet v2 13.58% words errors,23.5% speakers errors,18.6% average errors,$0.07 an hour
    5. AvenceraGranite Speech 13.79% words errors,23.5% speakers errors,18.7% average errors,$0.09 an hour
    6. AssemblyAI 12.80% words errors,33.2% speakers errors,23.0% average errors,$0.23 an hour
    7. ElevenLabs 13.66% words errors,39.8% speakers errors,26.8% average errors,$0.22 an hour

  • Podcasts and video

    Podcasts, YouTube videos and webinars

    Qwen3-ASR $0.09 / hour$0.10 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 7.99% words errors,16.9% speakers errors,12.5% average errors,$0.213 an hour
    2. AvenceraQwen3-ASR, recommended 7.22% words errors,23.5% speakers errors,15.4% average errors,$0.10 an hour
    3. AvenceraParakeet v3 7.99% words errors,23.5% speakers errors,15.8% average errors,$0.08 an hour
    4. AvenceraGranite Speech 8.14% words errors,23.5% speakers errors,15.8% average errors,$0.09 an hour
    5. AvenceraParakeet v2 8.15% words errors,23.5% speakers errors,15.8% average errors,$0.07 an hour
    6. AssemblyAI 7.62% words errors,33.2% speakers errors,20.4% average errors,$0.23 an hour
    7. ElevenLabs 7.63% words errors,39.8% speakers errors,23.7% average errors,$0.22 an hour

  • Earnings calls

    Investor calls and financial briefings

    Parakeet v3 $0.07 / hour$0.08 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 5.85% words errors,16.9% speakers errors,11.4% average errors,$0.213 an hour
    2. AvenceraQwen3-ASR 5.84% words errors,23.5% speakers errors,14.7% average errors,$0.10 an hour
    3. AvenceraParakeet v3, recommended 5.85% words errors,23.5% speakers errors,14.7% average errors,$0.08 an hour
    4. AvenceraGranite Speech 6.22% words errors,23.5% speakers errors,14.9% average errors,$0.09 an hour
    5. AvenceraParakeet v2 6.76% words errors,23.5% speakers errors,15.1% average errors,$0.07 an hour
    6. AssemblyAI 6.05% words errors,33.2% speakers errors,19.6% average errors,$0.23 an hour
    7. ElevenLabs 4.80% words errors,39.8% speakers errors,22.3% average errors,$0.22 an hour

  • Lectures and audiobooks

    Recorded courses, narration and other read-aloud audio

    Granite Speech $0.08 / hour$0.09 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 3.13% words errors,16.9% speakers errors,10.0% average errors,$0.213 an hour
    2. AvenceraGranite Speech, recommended 2.21% words errors,23.5% speakers errors,12.9% average errors,$0.09 an hour
    3. AvenceraParakeet v2 2.73% words errors,23.5% speakers errors,13.1% average errors,$0.07 an hour
    4. AvenceraQwen3-ASR 2.94% words errors,23.5% speakers errors,13.2% average errors,$0.10 an hour
    5. AvenceraParakeet v3 3.13% words errors,23.5% speakers errors,13.3% average errors,$0.08 an hour
    6. AssemblyAI 2.11% words errors,33.2% speakers errors,17.6% average errors,$0.23 an hour
    7. ElevenLabs 2.33% words errors,39.8% speakers errors,21.1% average errors,$0.22 an hour

  • Speeches

    Talks, keynotes and parliamentary sessions

    Qwen3-ASR $0.09 / hour$0.10 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 3.05% words errors,16.9% speakers errors,10.0% average errors,$0.213 an hour
    2. AvenceraQwen3-ASR, recommended 2.85% words errors,23.5% speakers errors,13.2% average errors,$0.10 an hour
    3. AvenceraParakeet v3 3.05% words errors,23.5% speakers errors,13.3% average errors,$0.08 an hour
    4. AvenceraGranite Speech 3.52% words errors,23.5% speakers errors,13.5% average errors,$0.09 an hour
    5. AvenceraParakeet v2 3.78% words errors,23.5% speakers errors,13.7% average errors,$0.07 an hour
    6. AssemblyAI 1.92% words errors,33.2% speakers errors,17.5% average errors,$0.23 an hour
    7. ElevenLabs 1.41% words errors,39.8% speakers errors,20.6% average errors,$0.22 an hour

  • Noisy recordings

    Field recordings, phone microphones and background noise

    Qwen3-ASR $0.09 / hour$0.10 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 11.52% words errors,16.9% speakers errors,14.2% average errors,$0.213 an hour
    2. AvenceraQwen3-ASR, recommended 9.93% words errors,23.5% speakers errors,16.7% average errors,$0.10 an hour
    3. AvenceraGranite Speech 10.22% words errors,23.5% speakers errors,16.9% average errors,$0.09 an hour
    4. AvenceraParakeet v2 11.42% words errors,23.5% speakers errors,17.5% average errors,$0.07 an hour
    5. AvenceraParakeet v3 11.52% words errors,23.5% speakers errors,17.5% average errors,$0.08 an hour
    6. AssemblyAI 9.08% words errors,33.2% speakers errors,21.1% average errors,$0.23 an hour
    7. ElevenLabs 9.20% words errors,39.8% speakers errors,24.5% average errors,$0.22 an hour

  • Indian English

    Calls and recordings from speakers in India

    Qwen3-ASR $0.09 / hour$0.10 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 4.14% words errors,16.9% speakers errors,10.5% average errors,$0.213 an hour
    2. AvenceraQwen3-ASR, recommended 3.50% words errors,23.5% speakers errors,13.5% average errors,$0.10 an hour
    3. AvenceraParakeet v2 3.89% words errors,23.5% speakers errors,13.7% average errors,$0.07 an hour
    4. AvenceraGranite Speech 3.94% words errors,23.5% speakers errors,13.7% average errors,$0.09 an hour
    5. AvenceraParakeet v3 4.14% words errors,23.5% speakers errors,13.8% average errors,$0.08 an hour
    6. AssemblyAI 3.49% words errors,33.2% speakers errors,18.3% average errors,$0.23 an hour
    7. ElevenLabs 3.32% words errors,39.8% speakers errors,21.6% average errors,$0.22 an hour

  • European languages

    German, French, Spanish, Italian, Portuguese, Dutch and 19 more

    Parakeet v3 $0.07 / hour$0.08 / hour with speaker labels

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 5.25% words errors,16.9% speakers errors,11.1% average errors,$0.213 an hour
    2. AvenceraParakeet v3, recommended 5.25% words errors,23.5% speakers errors,14.4% average errors,$0.08 an hour
    3. AvenceraQwen3-ASR 5.52% words errors,23.5% speakers errors,14.5% average errors,$0.10 an hour
    4. AssemblyAI 3.15% words errors,33.2% speakers errors,18.2% average errors,$0.23 an hour
    5. Speechmatics 4.35% words errors,32.1% speakers errors,18.2% average errors,$0.75 an hour
    6. Soniox 4.44% words errors,36.6% speakers errors,20.5% average errors,$0.10 an hour
    7. ElevenLabs 2.75% words errors,39.8% speakers errors,21.3% average errors,$0.22 an hour

  • Speaker labels only

    You already have a transcript and need to know who spoke when

    Speaker labels from community-1 $0.02 / hour

    More accurate speaker labels than Speechmatics, AssemblyAI, Soniox, ElevenLabs and Deepgram

    1. pyannoteAI 16.9% errors errors,$0.125 an hour
    2. Avenceracommunity-1, recommended 23.5% errors errors,$0.02 an hour
    3. Speechmatics 32.1% errors errors,$0.75 an hour
    4. AssemblyAI 33.2% errors errors,$0.23 an hour
    5. Soniox 36.6% errors errors,$0.10 an hour
    6. ElevenLabs 39.8% errors errors,$0.22 an hour
    7. Deepgram 46.1% errors errors,$0.258 an hour

Switch models with one line

Every model works with the OpenAI SDK. Change the model name and keep the rest of your code. Speaker labels are on by default, so every segment comes back with its speaker. Send speaker_labels as false for a plain transcript, a cent less an hour. Turn speaker labels off for single-speaker recordings or audio with one speaker per channel.

from openai import OpenAI

client = OpenAI(base_url="https://api.avencera.com/v1", api_key="YOUR_AVENCERA_KEY")

with open("audio.mp3", "rb") as audio:
    transcript = client.audio.transcriptions.create(
        model="avencera/qwen3-asr-1.7b",
        file=audio,
        # for a plain transcript without speakers, a cent less an hour
        # extra_body={"speaker_labels": False},
    )

for segment in transcript.segments:
    print(segment["speaker"], segment["text"])

See every price or compare providers one at a time.