EVA-Bench - Task Completion: leaderboard
Voice agents handle simulated phone calls from an audio user simulator in 213 enterprise scenarios (airline customer service, healthcare HR and IT service management), calling tools against a scenario database; EVA-Bench (Bogavelli et al., 2026). Task completion in percent: the share of trials whose final scenario-database state matches the ground truth exactly (the right tool calls with the right parameters and nothing extra), over five clean-audio trials per scenario, averaged equally over the three domains; higher is better.
Metric: Task completion (%; exact final scenario-database state); five clean-audio trials, 213 scenarios, equal mean over three domains. Source: servicenow.github.io. Saturation forecast: Rough model projection: around 2026. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Cartesia Managed Agents - GPT 5.5 | 92.59 | |
| 2 | Gemini 3.8 Live Extended Thinking (High) | 87.42 | |
| 3 | Grok Voice 3.0 | 85.76 | |
| 4 | Gemini 3.8 Live Extended Thinking (Minimal) | 85.74 | |
| 5 | Scribe v2.2 Realtime + GPT-5.4 + Eleven Flash v2 (ElevenAgents) | 79.94 | |
| 6 | Grok Voice 2.0 | 78.47 | |
| 7 | GPT Realtime 2.1 | 75.63 | |
| 8 | GPT Realtime 1.5 | 73.9 | |
| 9 | Scribe v2.2 Realtime + Gemini 3 Flash + TTS Conversational v3 (ElevenAgents) | 73.61 | |
| 10 | Gemini 3.8 Live | 71.09 |
Interactive version: theaggregate.ai/benchmark?slug=eva-bench-task-completion · How It Works · Data refreshed daily, snapshot 2026-11-12.