Gradium vs Inworld STT-1
Inworld STT-1 is ahead of Gradium on the Pipecat Dataset, time to first text and final-text delay. The Ocular Dataset is too close to call.
Last updated Methodology by Dileep Chagam
Accuracy vs speed
Each dot is one model. Bottom-left is best: fewer errors and a faster final transcript.
Which to pick
Inworld STT-1 is the better pick over Gradium on every measure that separates them: accuracy on clean, scripted audio (3.76 points lower), showing words while the caller is still talking (642ms sooner) and quick turn-taking once the caller stops (171ms sooner). For accuracy on real phone conversations, the gap is inside the margin of error, so neither is ahead.
The four metrics, side by side
- Gradium
- 6.48%17th
- Inworld STT-1
- 2.72%10th
Inworld STT-1, 3.76 points lower
- Gradium
- 4.45%12th
- Inworld STT-1
- 10.60%16th
Too close to call: inside the margin of error
- Gradium
- 1.71s12th
- Inworld STT-1
- 1.06s6th
Inworld STT-1, 642ms sooner
- Gradium
- 212ms11th
- Inworld STT-1
- 41ms2nd
Inworld STT-1, 171ms sooner
Lower is better on every metric. The small figure is each model's place in the whole field. What each metric measures.
- WER · Word error rate
- Share of words the model got wrong: replaced, missing or added. Lower is better.
- Pipecat Dataset
- 1,000 short public clips from Pipecat's STT benchmark dataset.
- Ocular Dataset
- 8 recordings from 4 real phone conversations, licensed from Ocular.
- TTFT · Time to first text
- How long after the audio starts streaming until the first words appear.
- TTFS · Final-text delay
- How long after the caller stops until the full transcript arrives. A voice agent waits for this before it replies.
First text and final text, at three percentiles
Gradium shows first text after 7.32s and final text after 280ms, and Inworld STT-1 shows first text after 1.07s and final text after 65ms, on a bad turn (P95) across the same streamed turns.
| Measure | Model | P50 | P90 | P95 |
|---|---|---|---|---|
| Time to first text | Gradium | 1.71s | 4.05s | 7.32s |
| Inworld STT-1 | 1.06s | 1.07s | 1.07s | |
| Final-text delay | Gradium | 212ms | 266ms | 280ms |
| Inworld STT-1 | 41ms | 60ms | 65ms |
- P50 / P90 / P95 · Typical, slow and slowest turns
- Half of turns are faster than P50. Only 1 in 10 is slower than P90, and 1 in 20 slower than P95.
How each ranks among all 17 models
Place among the 17 models on each metric. Lower is better on all four.
| Metric | Gradium | STT-1 |
|---|---|---|
| Pipecat Dataset WER | 6.48%17th of 17 | 2.72%10th of 17 |
| Ocular Dataset WER | 4.45%12th of 17 | 10.60%16th of 17 |
| Time to first text | 1.71s12th of 15 | 1.06s6th of 15 |
| Final-text delay | 212ms11th of 17 | 41ms2nd of 17 |
How much audio each model was tested on
| Model | Pipecat clips | Ocular recordings | Turns timed | Failed |
|---|---|---|---|---|
| Gradium | 1,000 | 8 | 198 | 0 |
| Inworld STT-1 | 1,000 | 8 | 206 | 0 |
Only clips with usable text count, so the two totals can differ. How WER is scored.
Related comparisons
- Gradium vs Deepgram Flux Multilingual
- Gradium vs Deepgram Flux English
- Gradium vs GPT-4o Transcribe
- Gradium vs Smallest Pulse
- Gradium vs Deepgram Nova-3
- Gradium vs GPT-4o Mini Transcribe
- Gradium vs Soniox STT-RT v5
- Gradium vs Speechmatics Linden
- Gradium vs Cartesia Ink 2
- Gradium vs GPT Realtime Whisper
- Gradium vs Reson8
- Gradium vs Google Chirp 3
- Gradium vs AssemblyAI Universal 3.6 Pro
- Inworld STT-1 vs Soniox STT-RT v5
- Inworld STT-1 vs Speechmatics Linden
- Inworld STT-1 vs Cartesia Ink 2
- Inworld STT-1 vs GPT Realtime Whisper
- Inworld STT-1 vs Reson8
- Inworld STT-1 vs GPT-4o Mini Transcribe
- Inworld STT-1 vs Google Chirp 3
- Inworld STT-1 vs Deepgram Nova-3
- Inworld STT-1 vs Smallest Pulse
- Inworld STT-1 vs AssemblyAI Universal 3.6 Pro
- Inworld STT-1 vs GPT-4o Transcribe
- Inworld STT-1 vs Deepgram Flux English
- Inworld STT-1 vs Deepgram Flux Multilingual
Frequently asked questions
Which is more accurate, Gradium or Inworld STT-1?
Inworld STT-1 is more accurate on the public clips. On the 1,000 public Pipecat clips, Gradium scores 6.48% and Inworld STT-1 2.72%, so Inworld STT-1 makes fewer errors by 3.76 points. On the eight licensed Ocular recordings, Gradium scores 4.45% and Inworld STT-1 10.60%, with likely ranges of 3.17 to 5.90% and 3.85 to 22.76%. The ranges overlap, so the difference is too close to call. Word error rate counts incorrect, missing, and extra words, so lower is better.
Is Gradium faster than Inworld STT-1?
No. Inworld STT-1 finishes the transcript 171ms sooner. Gradium has the final transcript 212ms after the speaker stops and Inworld STT-1 41ms, at the median. Both are measured on the same 206 streamed turns.
Is Gradium or Inworld STT-1 better for voice agents?
Inworld STT-1 finishes the transcript 171ms sooner, and the two are too close to call on real conversations. A voice agent cannot reply until the final transcript arrives, and the Ocular recordings are real conversations.
Do Gradium and Inworld STT-1 stream text while the speaker is talking?
Yes, both return partial text before the speaker stops. Gradium shows first text after 1.71s and Inworld STT-1 after 1.06s, at the median, so Inworld STT-1 is 642ms sooner.