VAmoS Pro Bench: leaderboard

Veris AI voice-agent simulation on the billing line of a regulated utility: one agent verifies the caller, explains bill increases, takes payments, opens or changes instalment plans and handles extensions over chained two-to-four-request calls, calm and angry simulated callers, and street, cafe and TV background noise. End-to-end task completion in percent, higher is better. Rows are voice stacks; scores are kept per underlying model (the speech-to-speech model, or the LLM of an STT, LLM, TTS cascade), best stack per model.

Metric: Task Completion (%). Source: veris.ai. Saturation forecast: Around February 2027. 15 models tracked. Updated 19 Nov 2026.

Top models

#ModelScoreOverall rank
1Phonic 157.5
2Grok Voice 2.0 Fast44.7
3Dialt Fast41.1
4Gemini 3.8 Live40.1
5GPT-Live 139.2
6GPT Realtime 2.137.9
7Dialt Smart37.8
8GPT-4.1 Mini36.4#799
9GPT Realtime 233.6
10Electron26.8
11Gemini 3.1 Flash Live25.6
12PolyAI Dialog-RSN-125.3
13Mistral Medium (2604)22.7#683
14Eleven Realtime v1 Mini19.6
15GPT-OSS-120B19#708

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=vamos-pro-bench · How It Works · Data refreshed daily, snapshot 2026-11-20.