GPT Realtime 2.1 vs Cascade

GPT Realtime 2.1 is more reliable, by 7.1 points. GPT Realtime 2.1 handled 23 of 70 real callers well on all three runs, and Cascade 18. A caller only counts when every run of the call scored 80 or more.

On reliability, yes: GPT Realtime 2.1 handles more callers well on every run than the cascade. GPT Realtime 2.1 is ahead on ELO rating, hearing score, turn taking score, speaking score, task handling score, reliability, calls handled well and calls fully resolved, and Cascade on calls where the agent went quiet and median latency. Both ran as the same agent, with the same prompt and tools.

GPT Realtime 2.1, by 21.9 points higher. GPT Realtime 2.1 fully resolved 34.3% of calls and Cascade 12.4%. A call counts only when every step the caller needed was done.

No. Cascade starts replying 0.10s sooner at the median. GPT Realtime 2.1 takes 2.16s at the median and 2.70s at p90; Cascade takes 2.06s and 2.49s. Both are measured on the call audio, from the caller finishing to the agent starting to reply.

Cascade, by 4.8 points. The agent went quiet for 10 seconds or more on 11.9% of GPT Realtime 2.1 calls and 7.1% of Cascade calls.

There is no verified per-minute rate for Cascade, so cost cannot be compared yet.

Cekura ran both as the whole agent with the same prompt, tools and phone connection. Both took the same 70 recorded real human callers on live calls, three times each. Each call is graded on 25 skills across Hearing, Turn Taking, Task Handling and Speaking, and passes at a score of 80 out of 100. Cascades run the same agent for reference and are never ranked.