Skip to content

Compare

Cohere Transcribe vs GPT-5 Nano vs Silero VAD

Price, speed, lengths, sizes, inputs and plan for Cohere Transcribe, GPT-5 Nano and Silero VAD, side by side, with example outputs from each.

Side by side

Cohere Transcribe vs GPT-5 Nano vs Silero VAD: what each model offers
FactCohere TranscribeGPT-5 NanoSilero VAD
Maker Cohere OpenAI Silero
Makes Speech to text Agent & text Speech to text
Best at Transcribe speech Ask and write Transcribe speech
Price per unit $0.000072 / second about $0.052 in / $0.42 out per 1M tokens $0.000011 / second
Typical run $0.000364 (5 s of output) about $0.000263 (1,000 input + 500 output tokens) $0.000053 (5 s of output)
Lengths Per run— — —
Longest run — — —
Takes Audio Text Audio
Success rate Last 30 days— — —
Median time Last 30 days— — —
Plan needed Pro Free Pro

A dash means we do not have that figure for the model. Success rate and median time cover the last 30 days and appear once a model has enough runs.

Example outputs

Questions

Which plan do I need for Cohere Transcribe, GPT-5 Nano and Silero VAD?
Cohere Transcribe: Needs the Pro plan or higher. GPT-5 Nano: Runs on any account, including the Free plan. Silero VAD: Needs the Pro plan or higher.
Who makes Cohere Transcribe, GPT-5 Nano and Silero VAD?
Cohere Transcribe is made by Cohere, GPT-5 Nano is made by OpenAI and Silero VAD is made by Silero.

Keep looking

How these figures are made

Success rate and time to result
Last 30 days of customer runs. Success rate = successful runs / (successful runs + runs that failed on our side or the model's side). Runs refused by a safety filter, runs with invalid settings or missing input, runs stopped by the customer and runs without enough credits are not counted as failures. Time to result = from submitting to the finished result, including any queue wait; we show the median. Figures are hidden until a model has 20 runs in the window.
Prices
List price of one unit on the model's default route, in credits and US dollars at 10,000 credits = $1. Plans give credits at their own effective rate; failed or refused runs are not charged.
Share
Each model’s share of the week’s successful runs of its type, as a band. We publish ranks and bands, never the number of runs.