Benchmark Almanac

Cekura Turn Bench

Reson8 Rumble-2 96% tops its leaderboard GPT-Realtime-2.1 84.7% leads the models we track
Not in model cards

Mean end-of-turn score across six line conditions, as published by Cekura.

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
16
Entries on its own leaderboard, 5 models we track

Reson8 Rumble-2 tops the leaderboard with 96%

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 GPT-Realtime-2.1, at 84.7%.

16 entries. Scroll sideways for the rest.

  1. 96.0 Reson8 Rumble-2
  2. 87.3 Higgs Realtime
  3. 85.5 Amazon Nova 2 Sonic
  4. 84.7 GPT-Realtime-2.1
  5. 84.2 GPT-Live-1
  6. 83.9 Gemini 3.8 Live Ext. Thinking
  7. 83.2 Phonic v1
  8. 82.8 Gemini 3.1 Flash Live
  9. 81.5 Gemini 3.8 Live
  10. 78.4 Deepslate Opal
  11. 77.7 Azure Realtime
  12. 77.2 GPT-Realtime-2.1 Mini
  13. 74.5 Grok Voice 3.0
  14. 73.7 Grok Voice 2.0
  15. 71.3 Smart Turn v3
  16. 50.3 Energy VAD
  1. 1 Reson8 Rumble-2 reference baseline 96%
  2. 2 Higgs Realtime Default voice 87.3%
  3. 3 Amazon Nova 2 Sonic Voice matthew; endpointing sensitivity MEDIUM 85.5%
  4. 4 GPT-Realtime-2.1 Default voice 84.7%
  5. 5 GPT-Live-1 Backend gpt-6-sol at reasoning effort low; voice marin 84.2%
  6. 6 Gemini 3.8 Live Ext. Thinking Thinking level HIGH (required by this model); default voice 83.9%
  7. 7 Phonic v1 Voice sabrina; intelligence level high 83.2%
  8. 8 Gemini 3.1 Flash Live Default voice; default thinking 82.8%
  9. 9 Gemini 3.8 Live Default voice; default thinking 81.5%
  10. 10 Deepslate Opal ElevenLabs voice 21m00Tcm4TlvDq8ikWAM 78.4%
  11. 11 Azure Realtime Voice ava (Azure native voice); input transcription azure-speech en-US 77.7%
  12. 12 GPT-Realtime-2.1 Mini Default voice 77.2%
  13. 13 Grok Voice 3.0 Default voice 74.5%
  14. 14 Grok Voice 2.0 Default voice 73.7%
  15. 15 Smart Turn v3 reference baseline 71.3%
  16. 16 Energy VAD reference baseline 50.3%

Source: Official Cekura Turn Bench leaderboard

#EntryScoreSettingDateSource
1 Reson8 Rumble-2 96% reference baseline Oct 2026 Official leaderboard
2 Higgs Realtime 87.3% Default voice Oct 2026 Official leaderboard
3 Amazon Nova 2 Sonic 85.5% Voice matthew; endpointing sensitivity MEDIUM Oct 2026 Official leaderboard
4 GPT-Realtime-2.1 84.7% Default voice Jul 2026 Official leaderboard
5 GPT-Live-1 84.2% Backend gpt-6-sol at reasoning effort low; voice marin Oct 2026 Official leaderboard
6 Gemini 3.8 Live Ext. Thinking 83.9% Thinking level HIGH (required by this model); default voice Oct 2026 Official leaderboard
7 Phonic v1 83.2% Voice sabrina; intelligence level high Oct 2026 Official leaderboard
8 Gemini 3.1 Flash Live 82.8% Default voice; default thinking Oct 2026 Official leaderboard
9 Gemini 3.8 Live 81.5% Default voice; default thinking Oct 2026 Official leaderboard
10 Deepslate Opal 78.4% ElevenLabs voice 21m00Tcm4TlvDq8ikWAM Oct 2026 Official leaderboard
11 Azure Realtime 77.7% Voice ava (Azure native voice); input transcription azure-speech en-US Oct 2026 Official leaderboard
12 GPT-Realtime-2.1 Mini 77.2% Default voice Jul 2026 Official leaderboard
13 Grok Voice 3.0 74.5% Default voice Oct 2026 Official leaderboard
14 Grok Voice 2.0 73.7% Default voice Jul 2026 Official leaderboard
15 Smart Turn v3 71.3% reference baseline Oct 2026 Official leaderboard
16 Energy VAD 50.3% reference baseline Oct 2026 Official leaderboard

Number one over time

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

OpenAI
Own leaderboard 1 leader
GPT-Realtime-2.1 84.7%

Source: Official Cekura Turn Bench 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
0% 20% 40% 60% 80% 100% Near the ceiling (85%) GPT-Realtime-2.1 0% 20% 40% 60% 80% 100% Near the ceiling (85%) GPT-Realtime-2.1

Source: Official Cekura Turn Bench 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 232 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%