vCX Benchmark

Retell AI scores the models a Retell voice agent can run on four suites of pass-or-fail checks: following role constraints and business rules, keeping answers grounded in the prompt, tools, and retrieved context, calling the right tool with valid arguments, and replying in a way that is natural, concise, and safe to speak aloud. Retell publishes each suite’s pass rate and their equal-weight mean, which the board ranks on, beside Retell’s price per minute and the median time live callers wait for a model’s first sentence. This board shows the 15 highest-scoring of the 44 models Retell scores on all four suites.

Last updated · Source leaderboard checked daily

Main Insights

Results

vCX Benchmark results, ranked by score (mean pass rate), highest first. Last updated .
RankModelProviderScore (mean pass rate)Cost (USD / task)Duration (seconds / task)
1Tell-1Retell AI94.5%$0.0350
2GPT-6.1 SolOpenAI93.4%$0.0641
3GPT-6 AstraOpenAI93.2%$0.321
4Gemini 4 ArgonGoogle92.6%$0.134
5GPT-6.1 TerraOpenAI92.4%$0.0321
6GPT-6 SolOpenAI92.3%$0.0641
7Claude Opus 5.5Anthropic92.2%$0.132
8GPT-5.5OpenAI91.9%$0.161
9Claude Sonnet 5.5Anthropic91.2%$0.0641
10grok-voice-3.0xAI91.2%$0.14—
11Gemini 3.8 FlashGoogle90.6%$0.0482
12Claude Opus 5Anthropic90.4%$0.161
13Muse Spark 1.3Meta90.1%——
14GPT-5.6 TerraOpenAI90.1%$0.0640
15GPT-5.4OpenAI89.6%$0.0801