Benchmark Almanac

Cekura Agent Performance

Phonic v1 90.2% tops its leaderboard Grok Voice 3.0 81.7% leads the models we track
Not in model cards

Scenario reliability across three repeated calls (Pass³).

Separate task board. No combined Cekura score exists. Costs use their explicitly named units.

Made by
Cekura
Released
July 2026
Kind
Audio understanding and voice
Level
Everyday
Scored
Graded by a model
Answers
Agent task
Tasks
Public, plus a private test set
Last model added
October 2026
Scores refreshed
November 12th, 2026
15
Entries on its own leaderboard, 5 models we track

Phonic v1 tops the leaderboard with 90.2%

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. 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 Grok Voice 3.0, at 81.7%.

15 entries. Scroll sideways for the rest.

  1. 90.2 Phonic v1
  2. 82.9 Cascade
  3. 81.7 Grok Voice 3.0
  4. 79.3 GPT-Realtime-2.1
  5. 75.6 GPT-Live 1 + GPT-6 Sol
  6. 70.7 Azure Realtime
  7. 68.3 Rime Prism
  8. 67.1 Grok Voice 2.0
  9. 62.2 Gemini 3.1 Flash Live Preview
  10. 58.5 Deepslate Opal
  11. 56.1 Higgs Realtime
  12. 52.4 Gemini 3.8 Live Extended Thinking
  13. 50.0 GPT-Realtime-2.1 Mini
  14. 50.0 Amazon Nova 2 Sonic
  15. 41.5 Gemini 3.8 Live
  1. 1 Phonic v1 Intelligence high 90.2%
  2. 2 Cascade No reasoning setting; reference baseline 82.9%
  3. 3 Grok Voice 3.0 Reasoning high 81.7%
  4. 4 GPT-Realtime-2.1 Reasoning high 79.3%
  5. 5 GPT-Live 1 + GPT-6 Sol Sol on low thinking 75.6%
  6. 6 Azure Realtime No reasoning setting 70.7%
  7. 7 Rime Prism No reasoning setting 68.3%
  8. 8 Grok Voice 2.0 Reasoning high 67.1%
  9. 9 Gemini 3.1 Flash Live Preview Thinking level high 62.2%
  10. 10 Deepslate Opal No reasoning setting 58.5%
  11. 11 Higgs Realtime No reasoning setting 56.1%
  12. 12 Gemini 3.8 Live Extended Thinking Thinking level high 52.4%
  13. 13 GPT-Realtime-2.1 Mini Reasoning high 50%
  14. 14 Amazon Nova 2 Sonic Endpointing medium 50%
  15. 15 Gemini 3.8 Live No thinking level 41.5%

Source: Official Cekura Agent Performance leaderboard

#EntryScoreSettingDateSource
1 Phonic v1 90.2% Intelligence high Nov 2026 Official leaderboard
2 Cascade 82.9% No reasoning setting; reference baseline Nov 2026 Official leaderboard
3 Grok Voice 3.0 81.7% Reasoning high Oct 2026 Official leaderboard
4 GPT-Realtime-2.1 79.3% Reasoning high Jul 2026 Official leaderboard
5 GPT-Live 1 + GPT-6 Sol 75.6% Sol on low thinking Nov 2026 Official leaderboard
6 Azure Realtime 70.7% No reasoning setting Nov 2026 Official leaderboard
7 Rime Prism 68.3% No reasoning setting Nov 2026 Official leaderboard
8 Grok Voice 2.0 67.1% Reasoning high Jul 2026 Official leaderboard
9 Gemini 3.1 Flash Live Preview 62.2% Thinking level high Nov 2026 Official leaderboard
10 Deepslate Opal 58.5% No reasoning setting Nov 2026 Official leaderboard
11 Higgs Realtime 56.1% No reasoning setting Nov 2026 Official leaderboard
12 Gemini 3.8 Live Extended Thinking 52.4% Thinking level high Sep 2026 Official leaderboard
13 GPT-Realtime-2.1 Mini 50% Reasoning high Jul 2026 Official leaderboard
14 Amazon Nova 2 Sonic 50% Endpointing medium Nov 2026 Official leaderboard
15 Gemini 3.8 Live 41.5% No thinking level Nov 2026 Official leaderboard

Reliability vs. cost per call

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

60% 80% 100% $0$0.05$0.1$0.15$0.2 Cost per call ($), lower is better Pass³ GPT-Realtime-2.1 (Reasoning high): Pass³ 79.3%, cost per call $0.17 GPT-Realtime-2.1 Mini (Reasoning high): Pass³ 50%, cost per call $0.04 Grok Voice 2.0 (Reasoning high): Pass³ 67.1%, cost per call $0.15 Grok Voice 3.0 (Reasoning high): Pass³ 81.7%, cost per call $0.14 Azure Realtime (No reasoning setting): Pass³ 70.7%, cost per call $0.17. Not in our model list Gemini 3.1 Flash Live Preview (Thinking level high): Pass³ 62.2%, cost per call $0.13. Not in our model list Phonic v1 (Intelligence high): Pass³ 90.2%, cost per call $0.12. Not in our model list GPT-Live 1 + GPT-6 Sol (Sol on low thinking): Pass³ 75.6%, cost per call $0.11. Not in our model list Deepslate Opal (No reasoning setting): Pass³ 58.5%, cost per call $0.05. Not in our model list Higgs Realtime (No reasoning setting): Pass³ 56.1%, cost per call $0.04. Not in our model list Phonic v1 (Intelligence high)GPT-Live 1 + GPT-6 Sol (Sol on low thinking)Deepslate Opal (No reasoning setting)Higgs Realtime (No reasoning setting) 60% 80% 100% $0$0.05$0.1$0.15$0.2 Cost per call ($), lower is better Pass³ GPT-Realtime-2.1 (Reasoning high): Pass³ 79.3%, cost per call $0.17 GPT-Realtime-2.1 Mini (Reasoning high): Pass³ 50%, cost per call $0.04 Grok Voice 2.0 (Reasoning high): Pass³ 67.1%, cost per call $0.15 Grok Voice 3.0 (Reasoning high): Pass³ 81.7%, cost per call $0.14 Azure Realtime (No reasoning setting): Pass³ 70.7%, cost per call $0.17. Not in our model list Gemini 3.1 Flash Live Preview (Thinking level high): Pass³ 62.2%, cost per call $0.13. Not in our model list Phonic v1 (Intelligence high): Pass³ 90.2%, cost per call $0.12. Not in our model list GPT-Live 1 + GPT-6 Sol (Sol on low thinking): Pass³ 75.6%, cost per call $0.11. Not in our model list Deepslate Opal (No reasoning setting): Pass³ 58.5%, cost per call $0.05. Not in our model list Higgs Realtime (No reasoning setting): Pass³ 56.1%, cost per call $0.04. Not in our model list Phonic v1 (Intelligence high)GPT-Live 1 + GPT-6 Sol (Sol on low thinking)Deepslate Opal (No reasoning setting)Higgs Realtime (No reasoning setting)
ModelSettingPass³Cost per call
Phonic v1 Intelligence high 90.2% $0.12
Grok Voice 3.0 Reasoning high 81.7% $0.14
GPT-Realtime-2.1 Reasoning high 79.3% $0.17
GPT-Live 1 + GPT-6 Sol Sol on low thinking 75.6% $0.11
Azure Realtime No reasoning setting 70.7% $0.17
Grok Voice 2.0 Reasoning high 67.1% $0.15
Gemini 3.1 Flash Live Preview Thinking level high 62.2% $0.13
Deepslate Opal No reasoning setting 58.5% $0.05
Higgs Realtime No reasoning setting 56.1% $0.04
GPT-Realtime-2.1 Mini Reasoning high 50% $0.04

2 models have held number one on the official leaderboard since its release in July 2026

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

OpenAI xAI
Own leaderboard 2 leaders
Grok Voice 3.0 81.7%

Source: Official Cekura Agent Performance leaderboard

Scores over time

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% Near the ceiling (85%) Grok Voice 3.0 GPT-Realtime-2.1 0% 20% 40% 60% 80% 100% Near the ceiling (85%) Grok Voice 3.0

Source: Official Cekura Agent Performance 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%