GPT Realtime 2.1 is ahead on reliability, success rate, agent responses, data accuracy, stalled calls and interruption, and Grok Voice Think Fast 2.0 on response time and cost.
Last updated Methodology by Dileep Chagam
Pick Grok Voice Think Fast 2.0 for fast replies (0.31s sooner) and lower cost ($0.005 cheaper). Pick GPT Realtime 2.1 for reliability across repeated calls (12.2 points higher), calls that pass every check (13.0 points higher), getting the agent's replies right (2.4 points higher), saving the caller's data correctly (13.1 points higher), fewer stalled calls (1.2 points lower) and handling interruptions (0.06 higher).
Scenarios that passed on all three runs.
GPT Realtime 2.1, 12.2 points higher
Calls that passed every check.
GPT Realtime 2.1, 13.0 points higher
Calls where the agent's replies reached the expected outcome.
GPT Realtime 2.1, 2.4 points higher
Calls where everything the agent saved matched exactly.
GPT Realtime 2.1, 13.1 points higher
Calls where the agent went silent for 10 seconds or more.
GPT Realtime 2.1, 1.2 points lower
How well it handled being talked over, 0 to 5.
GPT Realtime 2.1, 0.06 higher
Median time to start replying after the caller finishes.
Grok Think Fast 2.0, 0.31s sooner
One minute of call at the vendor's published rates.
Grok Think Fast 2.0, $0.005 cheaper
The small figure is the place among ranked models. The cascade is not ranked.
Grok Think Fast 2.0 1.62s, GPT Realtime 2.1 1.94s at the median.
GPT Realtime 2.1 is more reliable, by 12.2 points. Grok Voice Think Fast 2.0 passed 55 of 82 scenarios on all three runs, and GPT Realtime 2.1 passed 65. A scenario only counts when every run of it passed.
For a phone agent, GPT Realtime 2.1, since it gets more scenarios right on every run. GPT Realtime 2.1 is ahead on reliability, success rate, agent responses, data accuracy, stalled calls and interruption, and Grok Voice Think Fast 2.0 on response time and cost.
GPT Realtime 2.1, by 13.1 points. Grok Voice Think Fast 2.0 saved everything exactly right on 78.9% of the calls that save data, and GPT Realtime 2.1 on 92.0%. On Appointments, 97.1% against 95.9%. On Medicare, 31.8% against 81.8%. A call counts only when nothing is missing, wrong, extra or saved twice.
Yes. Grok Voice Think Fast 2.0 starts replying 0.31s sooner at the median. Grok Voice Think Fast 2.0 takes 1.62s at the median and 1.94s at p90; GPT Realtime 2.1 takes 1.94s and 2.46s. Both are measured on the call audio, from the caller finishing to the agent starting to reply.
GPT Realtime 2.1, by 0.06 on a 0 to 5 scale. Grok Voice Think Fast 2.0 scores 4.92 and GPT Realtime 2.1 4.98 for how well the agent handled being talked over.
GPT Realtime 2.1, by 1.2 points. The agent went silent for 10 seconds or more on 4.5% of Grok Voice Think Fast 2.0 calls and 3.3% of GPT Realtime 2.1 calls.
Grok Voice Think Fast 2.0 is $0.005 cheaper a minute. Grok Voice Think Fast 2.0 costs $0.080 a minute of call and GPT Realtime 2.1 $0.085, at each vendor's published rates.
Cekura ran both as the whole agent in the same open-source Pipecat pipeline, with the same prompt, tools and connection. A simulated caller worked through all 82 scenarios on live calls, three times each, 2,952 calls across the benchmark. A call passes only when the agent said the right things and saved the right data.
