Benchmark Almanac

VoxParity

Qwen3.8-Omni (file) 0.25 tops its leaderboard Gemini 3.7 Flash 0.23 leads the models we track
Not in model cards

Voice agents choose tool calls after listening to speech. Paired recordings keep the words fixed while changing an audible cue that affects the correct action.

The public development set contains 40 of the 183 scenarios. The leaderboard also includes held-out scenarios.

Made by
Bhavik Mangla
Released
September 2026
Kind
Audio understanding and voice
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text, Audio → Actions
Interaction
Single turn
Environment
APIs and tools
Refresh
One fixed set
Run by
A host
Strongest model tested
Gemini 3.8 Flash #24 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
23
Entries on its own leaderboard, 14 models we track

Qwen3.8-Omni (file) tops the leaderboard with 0.25 Gemini 3.7 Flash leads with 0.23 among models released by September 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 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 Gemini 3.7 Flash, at 0.23. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Qwen3.8-Omni (file) file 0.25
  2. 2 Gemini 3.7 Flash file 0.23
  3. 3 MiMo-V2.6 Pro file 0.20
  4. 4 Inkling (BaseTen upstream) file 0.20
  5. 5 StepAudio 3 file 0.18
  6. 6 Gemini 3.8 Flash file 0.18
  7. 7 MiMo-V2.6 Flash file 0.17
  8. 8 Gemini 2.5 native-audio Live realtime 0.15
  9. 9 MiMo-V2.5 file 0.12
  10. 10 Voxtral Small file 0.11
  11. 11 Qwen2.5-Omni-7B local 0.10
  12. 12 Gemini 3.1 Flash Live realtime 0.09
  13. 13 Gemini 3.8 Live realtime 0.07
  14. 14 Muse Spark 1.2 file 0.07
  15. 15 Qwen3-Omni-30B (local) local 0.05
  16. 16 GPT-Realtime-2.1 realtime 0.05
  17. 17 GPT-Realtime-2.1 Mini realtime 0.04
  18. 18 Gemma-4-E4B (local) local 0.04
  19. 19 Phi-4-multimodal local 0.00
  20. 20 Gemma-4-12B (local) local 0.00
  21. 21 Qwen-Audio-3.1 RT realtime -0.00
  22. 22 Grok Voice realtime -0.00
  23. 23 Nemotron 3 Nano Omni 30B-A3B file -0.10

Source: Official VoxParity leaderboard

#EntryScore ±SettingDateSource
1 Qwen3.8-Omni (file) 0.25 ±0.1 file Official leaderboard
2 Gemini 3.7 Flash 0.23 0.2 to 0.3 file Aug 2026 Official leaderboard
3 MiMo-V2.6 Pro 0.20 0.1 to 0.3 file Sep 2026 Official leaderboard
4 Inkling (BaseTen upstream) 0.20 ±0.1 file Official leaderboard
5 StepAudio 3 0.18 0.1 to 0.2 file Official leaderboard
6 Gemini 3.8 Flash 0.18 0.1 to 0.3 file Sep 2026 Official leaderboard
7 MiMo-V2.6 Flash 0.17 0.1 to 0.3 file Sep 2026 Official leaderboard
8 Gemini 2.5 native-audio Live 0.15 ±0.1 realtime Official leaderboard
9 MiMo-V2.5 0.12 ±0.1 file Apr 2026 Official leaderboard
10 Voxtral Small 0.11 ±0.1 file Jul 2025 Official leaderboard
11 Qwen2.5-Omni-7B 0.10 ±0.1 local Mar 2025 Official leaderboard
12 Gemini 3.1 Flash Live 0.09 ±0.1 realtime Mar 2026 Official leaderboard
13 Gemini 3.8 Live 0.07 ±0.1 realtime Sep 2026 Official leaderboard
14 Muse Spark 1.2 0.07 ±0.1 file Aug 2026 Official leaderboard
15 Qwen3-Omni-30B (local) 0.05 ±0.1 local Official leaderboard
16 GPT-Realtime-2.1 0.05 0 to 0.1 realtime Jul 2026 Official leaderboard
17 GPT-Realtime-2.1 Mini 0.04 0 to 0.1 realtime Jul 2026 Official leaderboard
18 Gemma-4-E4B (local) 0.04 ±0 local Official leaderboard
19 Phi-4-multimodal 0.00 ±0.1 local Feb 2025 Official leaderboard
20 Gemma-4-12B (local) 0.00 0 to 0 local Official leaderboard
21 Qwen-Audio-3.1 RT -0.00 ±0.1 realtime Official leaderboard
22 Grok Voice -0.00 -0.1 to 0 realtime Official leaderboard
23 Nemotron 3 Nano Omni 30B-A3B -0.10 ±0.1 file Apr 2026 Official leaderboard
#ModelScore ±SettingDateSource
1 Gemini 3.7 Flash 0.23 0.2 to 0.3 file Aug 2026 Official leaderboard
2 MiMo-V2.6 Pro 0.20 0.1 to 0.3 file Sep 2026 Official leaderboard
3 Gemini 3.8 Flash 0.18 0.1 to 0.3 file Sep 2026 Official leaderboard
4 MiMo-V2.6 Flash 0.17 0.1 to 0.3 file Sep 2026 Official leaderboard
5 MiMo-V2.5 0.12 ±0.1 file Apr 2026 Official leaderboard
6 Voxtral Small 0.11 ±0.1 file Jul 2025 Official leaderboard
7 Qwen2.5-Omni-7B 0.10 ±0.1 local Mar 2025 Official leaderboard
8 Gemini 3.1 Flash Live 0.09 ±0.1 realtime Mar 2026 Official leaderboard
9 Gemini 3.8 Live 0.07 ±0.1 realtime Sep 2026 Official leaderboard
10 Muse Spark 1.2 0.07 ±0.1 file Aug 2026 Official leaderboard
11 GPT-Realtime-2.1 0.05 0 to 0.1 realtime Jul 2026 Official leaderboard
12 GPT-Realtime-2.1 Mini 0.04 0 to 0.1 realtime Jul 2026 Official leaderboard
13 Phi-4-multimodal 0.00 ±0.1 local Feb 2025 Official leaderboard
14 Nemotron 3 Nano Omni 30B-A3B -0.10 ±0.1 file Apr 2026 Official leaderboard

Number one over time

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

Google DeepMind
Own leaderboard Not tested since Sep 2026

Source: Official VoxParity leaderboard

Scores over time

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

Plot with other benchmarks
On the benchmark’s own leaderboard
-0.10 0.00 0.10 0.20 0.30 2025 2026 -0.10 0.00 0.10 0.20 0.30 2025 2026

Source: Official VoxParity leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Audio and video

All 695 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%

Data audit

All 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (23 entries).
  • Pass Models scored 14 models we track, plus 9 other entries on its leaderboard.
  • Pass Newest score The newest is from September 2026, for MiMo-V2.6 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3.8 Flash (#24).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 14 rows for models we track and 9 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check