VAmoS Pro Bench
Voice agents handle several utility billing requests in one call. A model checks the spoken amounts and recorded actions against the requirements for each request.
Veris ranks 20 voice agents using 15 distinct language or speech models. The paper calls the 100-call test VAmoS Energy and reports 14 agents.
- Made by
- Veris AI
- Released
- September 2026
- Kind
- Customer service
- Level
- University
- Scored
- Graded by a model
- Answers
- Agent task
- Measures
- Capability
- Subject
- Economics and business
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools, A simulator
- Items from
- Real work and use, Generated
- Refresh
- One fixed set
- Run by
- A host
- Based on
- VAmoS Bench
- In views
- Agents, Audio and speech, Robotics, Law and finance
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
Phonic phonic-1 tops the leaderboard with 57.5% Gemini 3.8 Live leads with 40.1% 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 on its own with several settings, we keep its best. Every run is in the next chart. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. The best model we track is Gemini 3.8 Live, at 40.1%. Agents built around a model, such as grok voice + Grok Voice 2.0, are their own entries. We don't count them as the model's score. 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:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- Claude Opus 5.5
- and 16 more
- 57.5 Phonic phonic-1
- 44.7 grok voice + Grok Voice 2.0
- 41.1 Dialt Fast
- 40.1
Gemini 3.8 Live
- 39.2 GPT-Live-1
- 37.9 openai realtime 2.1 + GPT-Realtime-2.1
- 37.8 Dialt Smart
- 36.4 elevenlabs (ElevenLabs ASR + GPT-4.1 mini + eleven_flash_v2)
- 33.6 openai realtime + GPT-Realtime-2
- 29.4 livekit (Deepgram nova-3 + GPT-4.1 mini + ElevenLabs eleven_flash_v2)
- 26.8 Electron
- 25.6
Gemini 3.1 Flash Live
- 25.3 PolyAI Dialog-RSN-1
- 22.7 mistral-medium-2604
- 21.3 gradium (Gradium STT + GPT-4.1 mini + Gradium TTS)
- 20.5 vapi (Deepgram + GPT-4.1 mini + ElevenLabs)
- 19.8 deepgram (nova-3 + GPT-4.1 mini + aura-2)
- 19.6 Eleven Realtime v1 Mini
- 19.0 hugging face (Whisper large-v3 + gpt-oss-120b + Kokoro-82M)
- 17.3 pipecat (Deepgram nova-3 + GPT-4.1 mini + ElevenLabs eleven_flash_v2)
- 1 Phonic phonic-1 speech‑to‑speech 57.5%
- 2 grok voice + Grok Voice 2.0 speech‑to‑speech 44.7%
- 3 Dialt Fast speech‑to‑speech 41.1%
- 4
Gemini 3.8 Live speech‑to‑speech 40.1%
- 5 GPT-Live-1 speech‑to‑speech 39.2%
- 6 openai realtime 2.1 + GPT-Realtime-2.1 speech‑to‑speech 37.9%
- 7 Dialt Smart speech‑to‑speech 37.8%
- 8 elevenlabs (ElevenLabs ASR + GPT-4.1 mini + eleven_flash_v2) cascade 36.4%
- 9 openai realtime + GPT-Realtime-2 speech‑to‑speech 33.6%
- 10 livekit (Deepgram nova-3 + GPT-4.1 mini + ElevenLabs eleven_flash_v2) cascade 29.4%
- 11 Pulse + Electron + Lightning v3.1 Pro cascade 26.8%
- 12
Gemini 3.1 Flash Live speech‑to‑speech 25.6%
- 13 PolyAI Dialog-RSN-1 speech‑to‑speech 25.3%
- 14 Voxtral realtime + mistral-medium-2604 + Voxtral TTS cascade 22.7%
- 15 gradium (Gradium STT + GPT-4.1 mini + Gradium TTS) cascade 21.3%
- 16 vapi (Deepgram + GPT-4.1 mini + ElevenLabs) cascade 20.5%
- 17 deepgram (nova-3 + GPT-4.1 mini + aura-2) cascade 19.8%
- 18 Eleven Realtime v1 Mini speech‑to‑speech 19.6%
- 19 hugging face (Whisper large-v3 + gpt-oss-120b + Kokoro-82M) cascade 19%
- 20 pipecat (Deepgram nova-3 + GPT-4.1 mini + ElevenLabs eleven_flash_v2) cascade 17.3%
Source: Official VAmoS Pro Bench leaderboard
- 1
Gemini 3.8 Live speech‑to‑speech 40.1%
- 2 GPT-Live-1 speech‑to‑speech 39.2%
- 3
Gemini 3.1 Flash Live speech‑to‑speech 25.6%
Source: Official VAmoS Pro Bench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Phonic phonic-1 | 57.5% | speech‑to‑speech, phonic | Official leaderboard | |
| 2 | grok voice + Grok Voice 2.0 | 44.7% | speech‑to‑speech | Jul 2026 | Official leaderboard |
| 3 | Dialt Fast | 41.1% | speech‑to‑speech, dialt fast | Official leaderboard | |
| 4 | | 40.1% | speech‑to‑speech, gemini 3.8 live | Sep 2026 | Official leaderboard |
| 5 | GPT-Live-1 | 39.2% | speech‑to‑speech, gpt-live 1 | Jul 2026 | Official leaderboard |
| 6 | openai realtime 2.1 + GPT-Realtime-2.1 | 37.9% | speech‑to‑speech | Jul 2026 | Official leaderboard |
| 7 | Dialt Smart | 37.8% | speech‑to‑speech, dialt smart | Official leaderboard | |
| 8 | elevenlabs (ElevenLabs ASR + GPT-4.1 mini + eleven_flash_v2) | 36.4% | cascade | Apr 2025 | Official leaderboard |
| 9 | openai realtime + GPT-Realtime-2 | 33.6% | speech‑to‑speech | May 2026 | Official leaderboard |
| 10 | livekit (Deepgram nova-3 + GPT-4.1 mini + ElevenLabs eleven_flash_v2) | 29.4% | cascade | Apr 2025 | Official leaderboard |
| 11 | Pulse + Electron + Lightning v3.1 Pro | 26.8% | cascade, smallest ai | Official leaderboard | |
| 12 | | 25.6% | speech‑to‑speech, gemini 3.1 live | Mar 2026 | Official leaderboard |
| 13 | PolyAI Dialog-RSN-1 | 25.3% | speech‑to‑speech, polyai | Official leaderboard | |
| 14 | Voxtral realtime + mistral-medium-2604 + Voxtral TTS | 22.7% | cascade, mistral | Official leaderboard | |
| 15 | gradium (Gradium STT + GPT-4.1 mini + Gradium TTS) | 21.3% | cascade | Apr 2025 | Official leaderboard |
| 16 | vapi (Deepgram + GPT-4.1 mini + ElevenLabs) | 20.5% | cascade | Apr 2025 | Official leaderboard |
| 17 | deepgram (nova-3 + GPT-4.1 mini + aura-2) | 19.8% | cascade | Apr 2025 | Official leaderboard |
| 18 | Eleven Realtime v1 Mini | 19.6% | speech‑to‑speech, eleven realtime mini | Official leaderboard | |
| 19 | hugging face (Whisper large-v3 + gpt-oss-120b + Kokoro-82M) | 19% | cascade | Aug 2025 | Official leaderboard |
| 20 | pipecat (Deepgram nova-3 + GPT-4.1 mini + ElevenLabs eleven_flash_v2) | 17.3% | cascade | Apr 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 40.1% | speech‑to‑speech, gemini 3.8 live | Sep 2026 | Official leaderboard |
| 2 | GPT-Live-1 | 39.2% | speech‑to‑speech, gpt-live 1 | Jul 2026 | Official leaderboard |
| 3 | | 25.6% | speech‑to‑speech, gemini 3.1 live | Mar 2026 | Official leaderboard |
The best score, 44.7%, costs $0.21 a task
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official VAmoS Pro Bench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| grok voice + Grok Voice 2.0 | speech‑to‑speech | grok voice | 44.7% | $0.21 | |
| speech‑to‑speech | gemini 3.8 live | 40.1% | $0.19 | ||
| GPT-Live-1 | speech‑to‑speech | gpt-live 1 | 39.2% | $0.19 | |
| openai realtime 2.1 + GPT-Realtime-2.1 | speech‑to‑speech | openai realtime 2.1 | 37.9% | $0.63 | |
| elevenlabs (ElevenLabs ASR + GPT-4.1 mini + eleven_flash_v2) | cascade | elevenlabs | 36.4% | $0.27 | |
| openai realtime + GPT-Realtime-2 | speech‑to‑speech | openai realtime | 33.6% | $0.61 | |
| livekit (Deepgram nova-3 + GPT-4.1 mini + ElevenLabs eleven_flash_v2) | cascade | livekit | 29.4% | $0.08 | |
| speech‑to‑speech | gemini 3.1 live | 25.6% | $0.16 | ||
| gradium (Gradium STT + GPT-4.1 mini + Gradium TTS) | cascade | gradium | 21.3% | $0.12 | |
| vapi (Deepgram + GPT-4.1 mini + ElevenLabs) | cascade | vapi | 20.5% | $0.27 | |
| deepgram (nova-3 + GPT-4.1 mini + aura-2) | cascade | deepgram | 19.8% | $0.19 | |
| hugging face (Whisper large-v3 + gpt-oss-120b + Kokoro-82M) | cascade | hugging face | 19% | $0.05 | |
| pipecat (Deepgram nova-3 + GPT-4.1 mini + ElevenLabs eleven_flash_v2) | cascade | pipecat | 17.3% | $0.11 |
2 models have held number one on the official leaderboard since its release in September 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official VAmoS Pro Bench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official VAmoS Pro Bench 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.
Versions
More in Tool use
- AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state.
Gemini 4 Neon 56.3%
- MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks.
Gemini 4 Argon 90.7%
- BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes.
Qwen3.7-Max 75%
- τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state.
Qwen 3.8 Max 55.1%
- Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
- DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark
Catalog record 1 warning 1 pass
- Warning Task access We don’t know whether the tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (20 entries).
- Pass Models scored 3 models we track, plus 17 other entries on its leaderboard.
- Pass Newest score The newest is from September 2026, for Gemini 3.8 Live.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 40.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on veris.ai: 13 rows for models we track and 7 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.