Official provider results
Gemini 3.8 Audio model evaluation
Google’s September 15 model evaluation reports separate results for Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on spoken-agent task suites.
Source snapshots refreshed
Five measurement lanesVOICE / S2S
Reasoning, task completion, conversation dynamics, experience, and latency stay separate.
Benchmark profile
- Scope
- Live API voice-agent evaluations using ServiceNow EVA-Bench, τ-Voice, and Sierra τ³-Banking
- Primary metric
- Task success (higher is better)
- Owner
- Google DeepMind
- Available evidence
- Exact bar-chart labels in Google’s model evaluation PDF, plus figures stated in the prose of the launch announcement; EVA-Bench scatterplot points are not tabulated
What it measures
Task completion
Can it follow policy, use tools, and finish a workflow?
Conversation dynamics
Can it handle turns, interruptions, ambiguity, and state?
Voice experience
Is the exchange natural, robust, and responsive?
Available results
- Gemini 3.8 Live τ-Voice task success
- 30.1%
- Gemini 3.8 Live Extended Thinking τ-Voice task success
- 68.6%
- Gemini 3.8 Live Extended Thinking Sierra τ³-Banking
- 35.1%
Google’s comparison chart for the default Live model.
High-effort Extended Thinking run; do not attribute this result to the default Live model.
High-effort Live API run on the banking task suite.
Interpretation limit
Google’s PDF combines evaluations run by separate teams using different APIs and effort settings. The Extended Thinking task rows use high effort, and the EVA-Bench scatterplot labels are plotted without exact figures. These voice-protocol results do not enter weighted text-model scores, and the plotted audio-hour costs are not canonical API prices.