How accurate is Gradium?

Gradium transcribes the 1,000 public Pipecat clips at 6.48% word error rate and the eight licensed Ocular recordings at 4.45%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.

How fast is Gradium in a live conversation?

First text arrives after 1.71s at the median, and the transcript is final 212ms 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 Gradium 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. Gradium returned usable text for 1,000 of the public clips.

Bench

Book a demo

Gradium

Every model

Accuracy vs. speed

Lower and further left is better · Nemotron 3.5 Transcribe doesn't support streaming, so it is not shown

Best quadrantBest picks

Final-text delay P50 (ms)

Overall WER (%)

Gradium is 12th of 18 on final-text delay and 19th of 19 on the Pipecat Dataset.

5.78% Overall WER (18th) · 6.48% Pipecat WER (19th) · 4.45% Ocular WER (14th) · 1.71s First text (13th) · 212ms Final text (12th)Likely range, 95% confidence: Pipecat WER 5.93 to 7.06%, Ocular WER 3.17 to 5.90%No published hourly price. No per-hour rate: speech-to-text is paid in plan credits, 3 per second of audio. Source, checked Oct 8, 2026.

Last updated Sep 16, 2026Methodology by Dileep Chagam

Gradium against every other model

Lower is better on every measure. Hover a column for that model's full numbers.

NVIDIA Nemotron 3.5 Transcribe send no text until the speaker stops, so TTFT is not comparable.

Glossary: WER, TTFT, TTFS, P50

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.
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.

Where Gradium struggled

On real calls, dropped words are its biggest source of error. On clean clips, wrong words lead. Our guide explains how word error rate is measured.

64%

of its errors on the Ocular calls are words it left out.

Median across all 19 models: 49%

Mix of errors

Real calls (Ocular Dataset)

Clean clips (Pipecat Dataset)

TTFT and TTFS at P50, P90 and P95

Timed on the same 206 streamed Ocular turns as every other model. P90 and P95 show what a caller meets on a slow turn.

TTFS, final text

How long after the caller stops until the full transcript arrives. A voice agent waits for this before it replies.

20ms50ms100ms200ms500ms1s2s

Typical, P50: 212ms; 12th fastest of 18
Slow, P90: 266ms; 10th fastest of 18
Slowest, P95: 280ms; 10th fastest of 18

TTFT, first text

How long after the audio starts streaming until the first words appear.

200ms500ms1s2s5s10s20s

Typical, P50: 1.71s; 13th fastest of 16
Slow, P90: 4.05s; 13th fastest of 16
Slowest, P95: 7.32s; 13th fastest of 16

Every other model's typical turn, on the same axis.

1,000 Pipecat clips · 8 Ocular recordings · 198 turns timed · 0 failed requests

Gradium FAQ

Where does Gradium place in the benchmark?

Gradium places highest on final-text delay, 12th of 18 at 212ms. Its weakest placing is the Pipecat Dataset, 19th of 19 at 6.48%.

Gradium places highest on final-text delay, 12th of 18 at 212ms. Its weakest placing is the Pipecat Dataset, 19th of 19 at 6.48%.
Gradium transcribes the 1,000 public Pipecat clips at 6.48% word error rate and the eight licensed Ocular recordings at 4.45%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.
First text arrives after 1.71s at the median, and the transcript is final 212ms 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.
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. Gradium returned usable text for 1,000 of the public clips.

Building on Gradium? Test your agent the same way.

Cekura runs these simulated calls on your agent and scores every one.

Book a demo Start free