Benchmark Almanac

vCX Benchmark

tell-1 94.5% tops its leaderboard GPT-6.1 Sol 93.3% leads the models we track
Not in model cards

Models run inside a Retell voice agent on pass or fail checks for business rules, grounding, tool calls and speech-ready replies. Each suite reports its own pass rate.

Retell's board also compares latency and reliability across providers. 15 models in the Fireworks index.

Made by
Retell AI
Kind
Speech recognition and translation
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
44
Entries on its own leaderboard, 42 models we track

tell-1 tops the leaderboard with 94.5%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is GPT-6.1 Sol, at 93.3%.

44 entries. Scroll sideways for the rest.

  1. 94.5 tell-1
  2. 93.3 GPT-6.1 Sol
  3. 93.2 GPT-6 Astra
  4. 92.6 Gemini 4 Argon
  5. 92.4 GPT-6.1 Terra
  6. 92.3 GPT-6 Sol
  7. 92.2 Claude Opus 5.5
  8. 91.9 GPT-5.5
  9. 91.2 Claude Sonnet 5.5
  10. 91.2 Grok Voice 3.0
  11. 90.6 Gemini 3.8 Flash
  12. 90.4 Claude Opus 5
  13. 90.1 Muse Spark 1.3
  14. 90.1 GPT-5.6 Terra
  15. 89.6 GPT-5.4
  16. 89.5 Claude Sonnet 4.6
  17. 88.8 Gemini 3.7 Flash
  18. 88.6 GPT-Realtime-1.5
  19. 87.5 Claude Sonnet 5
  20. 87.0 Claude Sonnet 4.5
  21. 87.0 Gemini 3.6 Flash
  22. 86.9 GPT-5.6 Luna
  23. 86.0 gpt-realtime
  24. 85.4 Claude Haiku 5.5
  25. 84.6 GPT-4.1
  26. 84.0 GPT-Realtime-2.1
  27. 83.9 GPT-Realtime-2
  28. 83.9 Gemini 3 Flash
  29. 83.2 GPT-5.2
  30. 82.9 GPT-4.1 mini
  31. 82.8 Claude Haiku 4.5
  32. 82.7 Gemini 3.5 Flash-Lite
  33. 82.6 Gemini 3.1 Flash-Lite
  34. 82.3 GPT-5
  35. 81.7 Gemini 3.5 Flash
  36. 81.5 GPT-5.1
  37. 80.7 GPT-6 Luna
  38. 77.9 GPT-5.4 Mini
  39. 75.0 gpt-realtime-mini
  40. 75.0 GPT-5 mini
  41. 74.5 GPT-5.4 Nano
  42. 64.4 GPT-4.1 nano
  43. 63.6 GPT-Realtime-2.1 Mini
  44. 56.4 GPT-5 nano
  1. 1 tell-1 interleaved; standard 94.5%
  2. 2 GPT-6.1 Sol low; standard 93.3%
  3. 3 GPT-6 Astra low; standard 93.2%
  4. 4 Gemini 4 Argon low; standard 92.6%
  5. 5 GPT-6.1 Terra none; standard 92.4%
  6. 6 GPT-6 Sol none; standard 92.3%
  7. 7 Claude Opus 5.5 low; standard 92.2%
  8. 8 GPT-5.5 none; standard 91.9%
  9. 9 Claude Sonnet 5.5 none; standard 91.2%
  10. 10 Grok Voice 3.0 high; standard 91.2%
  11. 11 Gemini 3.8 Flash low; standard 90.6%
  12. 12 Claude Opus 5 none; standard 90.4%
  13. 13 Muse Spark 1.3 minimal; standard 90.1%
  14. 14 GPT-5.6 Terra none; standard 90.1%
  15. 15 GPT-5.4 none; standard 89.6%
  16. 16 Claude Sonnet 4.6 none; standard 89.5%
  17. 17 Gemini 3.7 Flash low; standard 88.8%
  18. 18 GPT-Realtime-1.5 none; standard 88.6%
  19. 19 Claude Sonnet 5 none; standard 87.5%
  20. 20 Claude Sonnet 4.5 none; standard 87%
  21. 21 Gemini 3.6 Flash minimal; standard 87%
  22. 22 GPT-5.6 Luna none; standard 86.9%
  23. 23 gpt-realtime none; standard 86%
  24. 24 Claude Haiku 5.5 none; standard 85.4%
  25. 25 GPT-4.1 none; standard 84.6%
  26. 26 GPT-Realtime-2.1 none; standard 84%
  27. 27 GPT-Realtime-2 none; standard 83.9%
  28. 28 Gemini 3 Flash minimal; standard 83.9%
  29. 29 GPT-5.2 none; standard 83.2%
  30. 30 GPT-4.1 mini none; standard 82.9%
  31. 31 Claude Haiku 4.5 none; standard 82.8%
  32. 32 Gemini 3.5 Flash-Lite minimal; standard 82.7%
  33. 33 Gemini 3.1 Flash-Lite minimal; standard 82.6%
  34. 34 GPT-5 minimal; standard 82.3%
  35. 35 Gemini 3.5 Flash minimal; standard 81.7%
  36. 36 GPT-5.1 none; standard 81.5%
  37. 37 GPT-6 Luna none; standard 80.7%
  38. 38 GPT-5.4 Mini none; standard 77.9%
  39. 39 gpt-realtime-mini none; standard 75%
  40. 40 GPT-5 mini minimal; standard 75%
  41. 41 GPT-5.4 Nano none; standard 74.5%
  42. 42 GPT-4.1 nano none; standard 64.4%
  43. 43 GPT-Realtime-2.1 Mini none; standard 63.6%
  44. 44 GPT-5 nano minimal; standard 56.4%

Source: Official vCX Benchmark leaderboard

#EntryScoreSettingDateSource
1 tell-1 94.5% interleaved; standard, Retell Official leaderboard
2 GPT-6.1 Sol 93.3% low; standard, Retell Sep 2026 Official leaderboard
3 GPT-6 Astra 93.2% low; standard, Retell Sep 2026 Official leaderboard
4 Gemini 4 Argon 92.6% low; standard, Retell Sep 2026 Official leaderboard
5 GPT-6.1 Terra 92.4% none; standard, Retell Oct 2026 Official leaderboard
6 GPT-6 Sol 92.3% none; standard, Retell Sep 2026 Official leaderboard
7 Claude Opus 5.5 92.2% low; standard, Retell Sep 2026 Official leaderboard
8 GPT-5.5 91.9% none; standard, Retell Apr 2026 Official leaderboard
9 Claude Sonnet 5.5 91.2% none; standard, Retell Sep 2026 Official leaderboard
10 Grok Voice 3.0 91.2% high; standard, Retell Oct 2026 Official leaderboard
11 Gemini 3.8 Flash 90.6% low; standard, Retell Sep 2026 Official leaderboard
12 Claude Opus 5 90.4% none; standard, Retell Jul 2026 Official leaderboard
13 Muse Spark 1.3 90.1% minimal; standard, Retell Sep 2026 Official leaderboard
14 GPT-5.6 Terra 90.1% none; standard, Retell Jul 2026 Official leaderboard
15 GPT-5.4 89.6% none; standard, Retell Mar 2026 Official leaderboard
16 Claude Sonnet 4.6 89.5% none; standard, Retell Feb 2026 Official leaderboard
17 Gemini 3.7 Flash 88.8% low; standard, Retell Aug 2026 Official leaderboard
18 GPT-Realtime-1.5 88.6% none; standard, Retell Feb 2026 Official leaderboard
19 Claude Sonnet 5 87.5% none; standard, Retell Jun 2026 Official leaderboard
20 Claude Sonnet 4.5 87% none; standard, Retell Sep 2025 Official leaderboard
21 Gemini 3.6 Flash 87% minimal; standard, Retell Jul 2026 Official leaderboard
22 GPT-5.6 Luna 86.9% none; standard, Retell Jul 2026 Official leaderboard
23 gpt-realtime 86% none; standard, Retell Aug 2025 Official leaderboard
24 Claude Haiku 5.5 85.4% none; standard, Retell Oct 2026 Official leaderboard
25 GPT-4.1 84.6% none; standard, Retell Apr 2025 Official leaderboard
26 GPT-Realtime-2.1 84% none; standard, Retell Jul 2026 Official leaderboard
27 GPT-Realtime-2 83.9% none; standard, Retell May 2026 Official leaderboard
28 Gemini 3 Flash 83.9% minimal; standard, Retell Dec 2025 Official leaderboard
29 GPT-5.2 83.2% none; standard, Retell Dec 2025 Official leaderboard
30 GPT-4.1 mini 82.9% none; standard, Retell Apr 2025 Official leaderboard
31 Claude Haiku 4.5 82.8% none; standard, Retell Oct 2025 Official leaderboard
32 Gemini 3.5 Flash-Lite 82.7% minimal; standard, Retell Jul 2026 Official leaderboard
33 Gemini 3.1 Flash-Lite 82.6% minimal; standard, Retell Mar 2026 Official leaderboard
34 GPT-5 82.3% minimal; standard, Retell Aug 2025 Official leaderboard
35 Gemini 3.5 Flash 81.7% minimal; standard, Retell May 2026 Official leaderboard
36 GPT-5.1 81.5% none; standard, Retell Nov 2025 Official leaderboard
37 GPT-6 Luna 80.7% none; standard, Retell Sep 2026 Official leaderboard
38 GPT-5.4 Mini 77.9% none; standard, Retell Mar 2026 Official leaderboard
39 gpt-realtime-mini 75% none; standard, Retell Official leaderboard
40 GPT-5 mini 75% minimal; standard, Retell Aug 2025 Official leaderboard
41 GPT-5.4 Nano 74.5% none; standard, Retell Mar 2026 Official leaderboard
42 GPT-4.1 nano 64.4% none; standard, Retell Apr 2025 Official leaderboard
43 GPT-Realtime-2.1 Mini 63.6% none; standard, Retell Jul 2026 Official leaderboard
44 GPT-5 nano 56.4% minimal; standard, Retell Aug 2025 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-6.1 Sol Retell
low; standa…
93.3%
GPT-6 Astra Retell
low; standa…
93.2%
GPT-6.1 Terra Retell
none; stand…
92.4%
GPT-6 Sol Retell
none; stand…
92.3%
GPT-5.5 Retell
none; stand…
91.9%
GPT-5.6 Terra Retell
none; stand…
90.1%
GPT-5.4 Retell
none; stand…
89.6%
GPT-5.6 Luna Retell
none; stand…
86.9%
GPT-4.1 Retell
none; stand…
84.6%
GPT-5.2 Retell
none; stand…
83.2%
GPT-4.1 mini Retell
none; stand…
82.9%
GPT-5 Retell
minimal; st…
82.3%

Source: Official vCX Benchmark leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.1 Sol low; standard Retell 93.3%
GPT-6.1 Sol low; fast Retell 93.3%
GPT-6.1 Sol low; ultrafast Retell 93.3%
GPT-6 Astra low; standard Retell 93.2%
GPT-6 Astra low; fast Retell 93.2%
Gemini 4 Argon low; standard Retell 92.6%
GPT-6.1 Terra none; standard Retell 92.4%
GPT-6.1 Terra none; fast Retell 92.4%
GPT-6 Sol none; standard Retell 92.3%
GPT-6 Sol none; fast Retell 92.3%
Claude Opus 5.5 low; standard Retell 92.2%
GPT-5.5 none; standard Retell 91.9%
GPT-5.5 none; fast Retell 91.9%
Claude Sonnet 5.5 none; standard Retell 91.2%
Grok Voice 3.0 high; standard Retell 91.2%
Gemini 3.8 Flash low; standard Retell 90.6%
Claude Opus 5 none; standard Retell 90.4%
Muse Spark 1.3 minimal; standard Retell 90.1%
GPT-5.6 Terra none; standard Retell 90.1%
GPT-5.6 Terra none; fast Retell 90.1%
GPT-5.4 none; standard Retell 89.6%
GPT-5.4 none; fast Retell 89.6%
Claude Sonnet 4.6 none; standard Retell 89.5%
Gemini 3.7 Flash low; standard Retell 88.8%
GPT-Realtime-1.5 none; standard Retell 88.6%
Claude Sonnet 5 none; standard Retell 87.5%
Claude Sonnet 4.5 none; standard Retell 87%
Gemini 3.6 Flash minimal; standard Retell 87%
GPT-5.6 Luna none; standard Retell 86.9%
GPT-5.6 Luna none; fast Retell 86.9%
gpt-realtime none; standard Retell 86%
Claude Haiku 5.5 none; standard Retell 85.4%
GPT-4.1 none; standard Retell 84.6%
GPT-4.1 none; fast Retell 84.6%
GPT-Realtime-2.1 none; standard Retell 84%
GPT-Realtime-2 none; standard Retell 83.9%
Gemini 3 Flash minimal; standard Retell 83.9%
GPT-5.2 none; standard Retell 83.2%
GPT-5.2 none; fast Retell 83.2%
GPT-4.1 mini none; standard Retell 82.9%
GPT-4.1 mini none; fast Retell 82.9%
Claude Haiku 4.5 none; standard Retell 82.8%
Gemini 3.5 Flash-Lite minimal; standard Retell 82.7%
Gemini 3.1 Flash-Lite minimal; standard Retell 82.6%
GPT-5 minimal; standard Retell 82.3%
GPT-5 minimal; fast Retell 82.3%
Gemini 3.5 Flash minimal; standard Retell 81.7%
GPT-5.1 none; standard Retell 81.5%
GPT-5.1 none; fast Retell 81.5%
GPT-6 Luna none; standard Retell 80.7%
GPT-6 Luna none; fast Retell 80.7%
GPT-5.4 Mini none; standard Retell 77.9%
GPT-5.4 Mini none; fast Retell 77.9%
GPT-5 mini minimal; standard Retell 75%
GPT-5 mini minimal; fast Retell 75%
GPT-5.4 Nano none; standard Retell 74.5%
GPT-4.1 nano none; standard Retell 64.4%
GPT-Realtime-2.1 Mini none; standard Retell 63.6%
GPT-5 nano minimal; standard Retell 56.4%

8 models have held number one on the official leaderboard since 2023

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI Anthropic
Own leaderboard Passed 85% in Aug 2025, by gpt-realtime
GPT-6.1 Sol 93.3%

Source: Official vCX Benchmark leaderboard

Gpt-realtime on the official leaderboard passed 85% in August 2025

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) GPT-6.1 Sol GPT-5.5 Claude Sonnet 4.5 GPT-4.1 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) GPT-6.1 Sol Claude Sonnet 4.5

Source: Official vCX Benchmark leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Audio and video

All 233 in Audio and video
  • Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
  • LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization. Gemini 4 Argon 91.7%
  • MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
  • MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
  • FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better. Phi-4-multimodal 4.00%
  • VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%