AssemblyAI Universal 3.5 Pro
AssemblyAI Universal 3.5 Pro is 2nd of 17 on the Pipecat Dataset and 10th of 17 on the Ocular Dataset.
Last updated Methodology by Dileep Chagam
- 1.93%
- Pipecat WER · 2nd
- 4.23%
- Ocular WER · 10th
- 489ms
- First text · 2nd
- 180ms
- Final text · 9th
Likely range, with 95% confidence: Pipecat WER 1.69 to 2.18%, Ocular WER 3.57 to 4.68%.
Accuracy vs speed
Each dot is one model. Bottom-left is best: fewer errors and a faster final transcript.
AssemblyAI Universal 3.5 Pro against every other model
Open on the metric AssemblyAI Universal 3.5 Pro places highest on, with the other three a tab away. Lower is better on all of them, so the order runs best first.
Word error rate on 1,000 clips from Pipecat's STT benchmark dataset. Lower is better.
| # | Model | Pipecat Dataset | clips | |
|---|---|---|---|---|
| 1 | AssemblyAI Universal 3.6 Pro | 1.77% | 1,000 | |
| 2 | AssemblyAI Universal 3.5 Pro | 1.93% | 1,000 | |
| 3 | Google Chirp 3 | 2.00% | 1,000 | |
| 4 | Reson8 | 2.10% | 1,000 | |
| 5 | GPT Realtime Whisper | 2.16% | 1,000 | |
| 6 | Cartesia Ink 2 | 2.39% | 1,000 | |
| 7 | Speechmatics Linden | 2.46% | 998 | |
| 8 | Soniox STT-RT v5 | 2.54% | 1,000 | |
| 9 | Google Chirp 2 | 2.59% | 1,000 | |
| 10 | Inworld STT-1 | 2.72% | 1,000 | |
| 11 | GPT-4o Mini Transcribe | 3.35% | 1,000 | |
| 12 | Deepgram Nova-3 | 3.46% | 1,000 | |
| 13 | Smallest Pulse | 3.66% | 1,000 | |
| 14 | GPT-4o Transcribe | 3.90% | 1,000 | |
| 15 | Deepgram Flux English | 3.90% | 999 | |
| 16 | Deepgram Flux Multilingual | 4.96% | 999 | |
| 17 | Gradium | 6.48% | 1,000 |
Counts are the clips each model returned usable text for, so they differ between models and between datasets.
- 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.
- Likely range · 95% range
- Where the model's true score probably falls, given how much audio was tested. Overlapping ranges are too close to call.
- 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
Measured on the same 206 streamed Ocular Dataset turns as every other model. P90 and P95 say what a caller meets on a bad turn, which a median hides.
| Measure | P50 | P90 | P95 |
|---|---|---|---|
| Time to first text | 489ms | 966ms | 1.08s |
| Final-text delay | 180ms | 532ms | 692ms |
- 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.
AssemblyAI Universal 3.5 Pro on the other two views
The same charts the benchmark page carries, with AssemblyAI Universal 3.5 Pro ringed. Consistency across the two datasets is one reading; how fast a model starts against how fast it commits is the other.
Public clips vs real calls
Each dot is one model. Bottom-left is best on both datasets.
First words vs final transcript
Each dot is one model. Bottom-left is fastest at both.
How much audio it was tested on
- 1,000
- Pipecat clips scored
- 8
- Ocular recordings scored
- 205
- turns timed
- 1
- requests that failed
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.
AssemblyAI Universal 3.5 Pro head to head
- vs AssemblyAI Universal 3.6 Pro
- vs Google Chirp 3
- vs Reson8
- vs GPT Realtime Whisper
- vs Cartesia Ink 2
- vs Speechmatics Linden
- vs Soniox STT-RT v5
- vs Google Chirp 2
- vs Inworld STT-1
- vs GPT-4o Mini Transcribe
- vs Deepgram Nova-3
- vs Smallest Pulse
- vs GPT-4o Transcribe
- vs Deepgram Flux English
- vs Deepgram Flux Multilingual
- vs Gradium
Frequently asked questions
Where does AssemblyAI Universal 3.5 Pro place in the benchmark?
AssemblyAI Universal 3.5 Pro places highest on the Pipecat Dataset, 2nd of 17 at 1.93%. Its weakest placing is the Ocular Dataset, 10th of 17 at 4.23%.
How accurate is AssemblyAI Universal 3.5 Pro?
AssemblyAI Universal 3.5 Pro transcribes the 1,000 public Pipecat clips at 1.93% word error rate and the eight licensed Ocular recordings at 4.23%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.
How fast is AssemblyAI Universal 3.5 Pro in a live conversation?
First text arrives after 489ms at the median, and the transcript is final 180ms after the speaker stops. Both are measured on the same 206 streamed turns as every other model, and the page shows P90 and P95 as well, which is what a caller meets on a bad turn.
How was AssemblyAI Universal 3.5 Pro measured?
Every model runs the same two datasets under the same conditions: 1,000 public clips from Pipecat's STT benchmark dataset for accuracy at scale, and eight licensed conversation recordings from Ocular for real speech. Latency comes from 206 streamed turns, using each provider's own supported finalization contract. AssemblyAI Universal 3.5 Pro returned usable text for 1,000 of the public clips.