VAmoS Bench
Voice agents answer simulated bank customers and use tools to manage credit cards. A model checks each call against its transcript and recorded actions.
Veris ranks 17 voice agents on 100 calls. The agents use ten distinct language or speech models. Eight agents use GPT-4.1 mini.
- Made by
- Veris AI
- Released
- July 2026
- Kind
- Customer service
- Level
- University
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- Kept private
- Measures
- Capability
- Risk
- Privacy
- Subject
- Economics and business
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools, A simulator
- Items from
- Generated
- Refresh
- New versions replace it
- Run by
- A host
- Occupations
- Office and administrative support
- In views
- Agents, Audio and speech, Robotics, Law and finance, Content safety
- Last model added
- March 2026
- Scores refreshed
- November 20th, 2026
pipecat + GPT-4.1 mini tops the leaderboard with 71% Gemini 3.1 Flash Live leads with 62.3%
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 on its own is Gemini 3.1 Flash Live, at 62.3%. Agents built around a model, such as pipecat + GPT-4.1 mini, 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.
- 71.0 pipecat + GPT-4.1 mini
- 70.3 livekit + GPT-4.1 mini
- 69.3 vapi + GPT-4.1 mini
- 69.3 grok voice + Grok Voice 2.0
- 67.3 elevenlabs + GPT-4.1 mini
- 67.0 gradium + GPT-4.1 mini
- 67.0 cartesia + GPT-4.1 mini
- 67.0 openai realtime + GPT-Realtime-2
- 65.3 hugging face + DeepSeek-V4-Flash
- 64.0 Gemini 2.5 Native Audio
- 64.0 deepgram + GPT-4.1 mini
- 62.3
Gemini 3.1 Flash Live
- 61.3 retell + GPT-4.1 mini
- 59.7 mistral-small
- 58.7 hermes-4-70b
- 51.3 openai realtime mini + GPT-Realtime-2.1 Mini
- 43.0 nemotron + Nemotron 3 Nano 30B-A3B
- 1 pipecat + GPT-4.1 mini STT: Deepgram nova-3-general / TTS: ElevenLabs 71%
- 2 livekit + GPT-4.1 mini STT: Deepgram nova-3-general / TTS: ElevenLabs 70.3%
- 3 vapi + GPT-4.1 mini STT: Deepgram nova-3 / TTS: ElevenLabs 69.3%
- 4 grok voice + Grok Voice 2.0 69.3%
- 5 elevenlabs + GPT-4.1 mini STT: Scribe / TTS: eleven_flash_v2 67.3%
- 6 gradium + GPT-4.1 mini STT: Gradium STT / TTS: Gradium TTS 67%
- 7 cartesia + GPT-4.1 mini STT: Ink / TTS: Sonic 67%
- 8 openai realtime + GPT-Realtime-2 67%
- 9 hugging face + DeepSeek-V4-Flash STT: Whisper large-v3-turbo / TTS: Kokoro-82M 65.3%
- 10 Gemini 2.5 Native Audio 64%
- 11 deepgram + GPT-4.1 mini STT: nova-3 / TTS: aura-2-thalia-en 64%
- 12
Gemini 3.1 Flash Live 62.3%
- 13 retell + GPT-4.1 mini STT: Vendor ASR / TTS: ElevenLabs 61.3%
- 14 mistral-small STT: Voxtral Realtime / TTS: Voxtral TTS 59.7%
- 15 hermes-4-70b STT: Deepgram nova-3 / TTS: ElevenLabs 58.7%
- 16 openai realtime mini + GPT-Realtime-2.1 Mini 51.3%
- 17 nemotron + Nemotron 3 Nano 30B-A3B STT: Nemotron / TTS: Magpie 43%
Source: Official VAmoS Bench leaderboard
Source: Official VAmoS Bench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | pipecat + GPT-4.1 mini | 71% | STT: Deepgram nova-3-general / TTS: ElevenLabs | Apr 2025 | Official leaderboard |
| 2 | livekit + GPT-4.1 mini | 70.3% | STT: Deepgram nova-3-general / TTS: ElevenLabs | Apr 2025 | Official leaderboard |
| 3 | vapi + GPT-4.1 mini | 69.3% | STT: Deepgram nova-3 / TTS: ElevenLabs | Apr 2025 | Official leaderboard |
| 4 | grok voice + Grok Voice 2.0 | 69.3% | Jul 2026 | Official leaderboard | |
| 5 | elevenlabs + GPT-4.1 mini | 67.3% | STT: Scribe / TTS: eleven_flash_v2 | Apr 2025 | Official leaderboard |
| 6 | gradium + GPT-4.1 mini | 67% | STT: Gradium STT / TTS: Gradium TTS | Apr 2025 | Official leaderboard |
| 7 | cartesia + GPT-4.1 mini | 67% | STT: Ink / TTS: Sonic | Apr 2025 | Official leaderboard |
| 8 | openai realtime + GPT-Realtime-2 | 67% | May 2026 | Official leaderboard | |
| 9 | hugging face + DeepSeek-V4-Flash | 65.3% | STT: Whisper large-v3-turbo / TTS: Kokoro-82M | Apr 2026 | Official leaderboard |
| 10 | Gemini 2.5 Native Audio | 64% | gemini 2.5 native | Official leaderboard | |
| 11 | deepgram + GPT-4.1 mini | 64% | STT: nova-3 / TTS: aura-2-thalia-en | Apr 2025 | Official leaderboard |
| 12 | | 62.3% | gemini 3.1 live | Mar 2026 | Official leaderboard |
| 13 | retell + GPT-4.1 mini | 61.3% | STT: Vendor ASR / TTS: ElevenLabs | Apr 2025 | Official leaderboard |
| 14 | mistral-small | 59.7% | STT: Voxtral Realtime / TTS: Voxtral TTS, mistral small | Official leaderboard | |
| 15 | hermes-4-70b | 58.7% | STT: Deepgram nova-3 / TTS: ElevenLabs, hermes | Official leaderboard | |
| 16 | openai realtime mini + GPT-Realtime-2.1 Mini | 51.3% | Jul 2026 | Official leaderboard | |
| 17 | nemotron + Nemotron 3 Nano 30B-A3B | 43% | STT: Nemotron / TTS: Magpie | Dec 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 62.3% | gemini 3.1 live | Mar 2026 | Official leaderboard |
The best score, 71%, costs $0.04 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 Bench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| pipecat + GPT-4.1 mini | STT: Deepgram nova-3-general / TTS: ElevenLabs | pipecat | 71% | $0.04 | |
| livekit + GPT-4.1 mini | STT: Deepgram nova-3-general / TTS: ElevenLabs | livekit | 70.3% | $0.05 | |
| vapi + GPT-4.1 mini | STT: Deepgram nova-3 / TTS: ElevenLabs | vapi | 69.3% | $0.14 | |
| grok voice + Grok Voice 2.0 | grok voice | 69.3% | $0.13 | ||
| elevenlabs + GPT-4.1 mini | STT: Scribe / TTS: eleven_flash_v2 | elevenlabs | 67.3% | $0.11 | |
| gradium + GPT-4.1 mini | STT: Gradium STT / TTS: Gradium TTS | gradium | 67% | $0.07 | |
| cartesia + GPT-4.1 mini | STT: Ink / TTS: Sonic | cartesia | 67% | $0.15 | |
| openai realtime + GPT-Realtime-2 | openai realtime | 67% | $0.07 | ||
| hugging face + DeepSeek-V4-Flash | STT: Whisper large-v3-turbo / TTS: Kokoro-82M | hugging face | 65.3% | $0.15 | |
| deepgram + GPT-4.1 mini | STT: nova-3 / TTS: aura-2-thalia-en | deepgram | 64% | $0.11 | |
| gemini 3.1 live | 62.3% | $0.02 | |||
| retell + GPT-4.1 mini | STT: Vendor ASR / TTS: ElevenLabs | retell | 61.3% | $0.21 | |
| openai realtime mini + GPT-Realtime-2.1 Mini | openai realtime mini | 51.3% | $0.03 | ||
| nemotron + Nemotron 3 Nano 30B-A3B | STT: Nemotron / TTS: Magpie | nemotron | 43% |
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 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
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access The tasks are private.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (17 entries).
- Pass Models scored 1 model we track, plus 16 other entries on its leaderboard.
- Note Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash Live. That’s expected for a test replaced by VAmoS Pro Bench.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 62.3%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on veris.ai: 14 rows for models we track and 3 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.