About Cekura Bench

Cekura builds testing and monitoring tools for voice agents, and runs these benchmarks itself. Because we sell into the same market we measure, every benchmark is built so that you can check it rather than take our word for it.

Last updated

How we keep it fair

  • Same test for everyone

    Every model or platform in a benchmark runs the same scenarios, recordings, prompts, tools and scoring. Only the thing being compared changes.

  • Rankings come from the measurements

    Places are read straight off the measured results. Ties are shown as ties, and figures that can't be compared fairly are marked rather than ranked.

  • You can check the work

    Test cases, agent definitions and benchmark code are public on GitHub, and the speech-to-speech and workflow pages link to the recorded calls behind their numbers. The licensed Ocular recordings are the one exception.

  • Mistakes get fixed in the open

    If a figure looks wrong, open an issue on the repository. We rerun or correct it, the page's last updated date changes with the fix, and the change is logged in the changelog.

The conflict-of-interest disclosure, advisory board, vendor pre-review policy, versioning rules and changelog are on the governance page.

Who runs each benchmark

The same checks power our voice agent testing.

Reuse and cite

Results are published under CC BY 4.0, so you can reuse them with credit. The data behind the speech benchmarks can be downloaded as JSON.

Citation

Cekura Bench (2026). Voice AI Benchmarks. Cekura. https://benchmarks.cekura.ai

BibTeX
@misc{cekurabench2026,
  title  = {Cekura Bench: Voice AI Benchmarks},
  author = {Dileep Chagam and Luis Ojeda and Shashij Gupta},
  year   = {2026},
  url    = {https://benchmarks.cekura.ai},
  note   = {Cekura}
}

Test your voice agent the same way.

Cekura runs these simulated calls on your agent and scores every one.