AssemblyAI Universal 3.5 Pro vs AssemblyAI Universal 3.6 Pro
AssemblyAI Universal 3.6 Pro is ahead on the Pipecat Dataset, time to first text and final-text delay, and AssemblyAI Universal 3.5 Pro on the Ocular Dataset.
Quality and speed frontier
Every model's error rate on the public clips, against how long its transcript takes to settle.
Where Universal 3.6 Pro improved
Same clips, same turns, both releases.
Final text, median
49% sooner- Universal 3.5 Pro
- 180ms
- Universal 3.6 Pro
- 91ms
Time from the speaker stopping to the final transcript.
Final text, P95
66% sooner- Universal 3.5 Pro
- 692ms
- Universal 3.6 Pro
- 232ms
19 turns in 20 are final within this.
Turns over 500ms to final text
none left- Universal 3.5 Pro
- 25
- Universal 3.6 Pro
- 0
Out of the same 205 turns.
Word errors, Pipecat Dataset
38 fewer- Universal 3.5 Pro
- 460
- Universal 3.6 Pro
- 422
On the same 1,000 clips.
First text, P95
11% sooner- Universal 3.5 Pro
- 1.08s
- Universal 3.6 Pro
- 967ms
19 turns in 20 show first words within this.
The four metrics, side by side
Word error rate on 1,000 clips from Pipecat's STT benchmark dataset.
- AssemblyAI Universal 3.5 Pro
- 1.93%2nd
- AssemblyAI Universal 3.6 Pro
- 1.77%1st
AssemblyAI Universal 3.6 Pro, 0.16 points lower
Word error rate on eight licensed recordings provided by Ocular.
- AssemblyAI Universal 3.5 Pro
- 4.23%10th
- AssemblyAI Universal 3.6 Pro
- 4.50%14th
AssemblyAI Universal 3.5 Pro, 0.27 points lower
Median delay before the first words arrive, from the first audio packet.
- AssemblyAI Universal 3.5 Pro
- 489ms2nd
- AssemblyAI Universal 3.6 Pro
- 485ms1st
AssemblyAI Universal 3.6 Pro, 4ms sooner
Median delay after the speaker stops before the transcript is final.
- AssemblyAI Universal 3.5 Pro
- 180ms9th
- AssemblyAI Universal 3.6 Pro
- 91ms3rd
AssemblyAI Universal 3.6 Pro, 89ms sooner
Lower is better on every metric. The small figure is each model's place in the whole field.
First text and final text, at three percentiles
Both measured on the same 206 streamed Ocular Dataset turns. P90 and P95 say what a caller meets on a bad turn, which a median hides.
| Measure | Model | P50 | P90 | P95 |
|---|---|---|---|---|
| Time to first text | AssemblyAI Universal 3.5 Pro | 489ms | 966ms | 1.08s |
| AssemblyAI Universal 3.6 Pro | 485ms | 966ms | 967ms | |
| Final-text delay | AssemblyAI Universal 3.5 Pro | 180ms | 532ms | 692ms |
| AssemblyAI Universal 3.6 Pro | 91ms | 210ms | 232ms |
Both against every other model
Open on the metric where the two sit furthest apart, with the other three a tab away.
Final-text delay after the speaker stops. Lower is better.
| # | Model | Final text | turns | |
|---|---|---|---|---|
| 1 | Soniox STT-RT v5 | 38ms | 200 | |
| 2 | Inworld STT-1 | 41ms | 206 | |
| 3 | AssemblyAI Universal 3.6 Pro | 91ms | 206 | |
| 4 | Deepgram Nova-3 | 103ms | 202 | |
| 5 | Deepgram Flux English | 108ms | 199 | |
| 6 | Deepgram Flux Multilingual | 114ms | 175 | |
| 7 | Cartesia Ink 2 | 123ms | 198 | |
| 8 | Speechmatics Linden | 147ms | 194 | |
| 9 | AssemblyAI Universal 3.5 Pro | 180ms | 205 | |
| 10 | Smallest Pulse | 201ms | 206 | |
| 11 | Gradium | 212ms | 198 | |
| 12 | Reson8 | 235ms | 201 | |
| 13 | Google Chirp 3 | 475ms | 206 | |
| 14 | GPT Realtime Whisper | 543ms | 206 | |
| 15 | Google Chirp 2 | 568ms | 205 | |
| 16 | GPT-4o Mini Transcribe | 630ms | 206 | |
| 17 | GPT-4o Transcribe | 652ms | 206 |
Median across the same 206 Ocular Dataset turns. Providers finalize differently, so compare models with similar finalization contracts before treating small gaps as meaningful.
Both on the other two views
Pipecat vs Ocular accuracy
Every model's error rate on the public clips, against the same model on real conversations.
Transcription response time
How soon each model returns its first words, against how soon its transcript is final.
What the numbers rest on
| Model | Pipecat clips | Ocular recordings | Turns timed | Failed |
|---|---|---|---|---|
| AssemblyAI Universal 3.5 Pro | 1,000 | 8 | 205 | 1 |
| AssemblyAI Universal 3.6 Pro | 1,000 | 8 | 206 | 0 |
A clip counts only where the model returned usable text, so counts differ between models. Word error rate is measured the same way throughout: incorrect, missing, and extra words against the reference transcript.
Frequently asked questions
Is AssemblyAI Universal 3.5 Pro or AssemblyAI Universal 3.6 Pro more accurate?
On the 1,000 public Pipecat clips, AssemblyAI Universal 3.5 Pro scores 1.93% and AssemblyAI Universal 3.6 Pro 1.77%, so AssemblyAI Universal 3.6 Pro makes fewer errors by 0.16 points. On the eight licensed Ocular recordings, AssemblyAI Universal 3.5 Pro scores 4.23% and AssemblyAI Universal 3.6 Pro 4.50%, so AssemblyAI Universal 3.5 Pro makes fewer errors by 0.27 points. The two datasets disagree, so the better choice depends on whether your audio looks more like clean public clips or real conversations.
Which is faster, AssemblyAI Universal 3.5 Pro or AssemblyAI Universal 3.6 Pro?
AssemblyAI Universal 3.6 Pro returns first text 4ms sooner, at 485ms against 489ms at the median. AssemblyAI Universal 3.6 Pro has the final transcript 89ms sooner, 91ms after the speaker stops against 180ms. Both are measured on the same 206 streamed turns.
How were AssemblyAI Universal 3.5 Pro and AssemblyAI Universal 3.6 Pro compared?
Both ran the same two datasets under the same conditions as every other model in the benchmark: 1,000 public clips from Pipecat's STT benchmark dataset and eight licensed conversation recordings from Ocular. AssemblyAI Universal 3.5 Pro returned usable text for 1,000 of the public clips and AssemblyAI Universal 3.6 Pro for 1,000. Latency uses each provider's own supported finalization contract.