Vendor page

Parallelism and Latency (Smallest AI self-hosted Pulse STT)

Smallest AI, n.d.

Read on October 19, 2026

What it is: Smallest AI's developer documentation with throughput figures for its Pulse and Pulse Pro speech-to-text models run on a single self-hosted GPU. For Pulse Pro it gives RTFx for a single two-hour file on an NVIDIA L40S, with and without word timestamps, and an effective RTFx for batched audio.
How much weight it carries: The vendor's own internal benchmark on one GPU host, so it is a vendor figure rather than an independent test. The page says the numbers are steady-state, single-GPU, and should be re-benchmarked on the buyer's hardware. No publication date is shown.

Facts taken from this source

Smallest AI reports an effective real-time factor of about 1,000 for Pulse Pro on one NVIDIA L40S GPU when audio is sent in batched 50-second chunks without word timestamps.
“Sustained throughput on L40S (batched, 50-second chunks) Mode RPS (requests / second) Effective RTFx No word timestamps 19–21 ~1,000×” (Parallelism and Latency, Pulse Pro, Sustained throughput on L40S)
Smallest AI reports that Pulse Pro transcribes a single two-hour audio file at about 200 times real time on one NVIDIA L40S GPU when word timestamps are on, in about 36 seconds.
“Long-form audio on L40S (single 2 hr file) Mode Throughput (RTFx) 2 hr file latency No word timestamps 250–300× ~24–29 sec With word timestamps ~200× ~36 sec” (Parallelism and Latency, Pulse Pro, Long-form audio on L40S)
Smallest AI reports that Pulse Pro transcribes a single two-hour audio file at 250 to 300 times real time on one NVIDIA L40S GPU when word timestamps are off, so about 275 times on average (midpoint).
“Long-form audio on L40S (single 2 hr file) Mode Throughput (RTFx) 2 hr file latency No word timestamps 250–300× ~24–29 sec” (Parallelism and Latency, Pulse Pro, Long-form audio on L40S)

Vendor page

Smallest AI, n.d.

Read on October 19, 2026

What it is: Smallest AI's developer documentation with throughput figures for its Pulse and Pulse Pro speech-to-text models run on a single self-hosted GPU. For Pulse Pro it gives RTFx for a single two-hour file on an NVIDIA L40S, with and without word timestamps, and an effective RTFx for batched audio.
How much weight it carries: The vendor's own internal benchmark on one GPU host, so it is a vendor figure rather than an independent test. The page says the numbers are steady-state, single-GPU, and should be re-benchmarked on the buyer's hardware. No publication date is shown.

Smallest AI reports an effective real-time factor of about 1,000 for Pulse Pro on one NVIDIA L40S GPU when audio is sent in batched 50-second chunks without word timestamps.
“Sustained throughput on L40S (batched, 50-second chunks) Mode RPS (requests / second) Effective RTFx No word timestamps 19–21 ~1,000×” (Parallelism and Latency, Pulse Pro, Sustained throughput on L40S)
Smallest AI reports that Pulse Pro transcribes a single two-hour audio file at about 200 times real time on one NVIDIA L40S GPU when word timestamps are on, in about 36 seconds.
“Long-form audio on L40S (single 2 hr file) Mode Throughput (RTFx) 2 hr file latency No word timestamps 250–300× ~24–29 sec With word timestamps ~200× ~36 sec” (Parallelism and Latency, Pulse Pro, Long-form audio on L40S)
Smallest AI reports that Pulse Pro transcribes a single two-hour audio file at 250 to 300 times real time on one NVIDIA L40S GPU when word timestamps are off, so about 275 times on average (midpoint).
“Long-form audio on L40S (single 2 hr file) Mode Throughput (RTFx) 2 hr file latency No word timestamps 250–300× ~24–29 sec” (Parallelism and Latency, Pulse Pro, Long-form audio on L40S)