Gemini 3.1 Flash Live scores 95.9% on Big Bench Audio, Google's fastest voice model
Google has released Gemini 3.1 Flash Live, its new voice and audio AI model, scoring 95.9% on the Big Bench Audio Benchmark at high thinking levels—second only to Step-Audio R1.1 Realtime at 97.0%. Response times range from 0.96 seconds at minimal thinking to 2.98 seconds at high thinking, with pricing held at $0.35 per hour of audio input and $1.40 per hour of audio output.
Gemini 3.1 Flash Live at a glance
Google Releases Gemini 3.1 Flash Live Voice Model
Google has unveiled Gemini 3.1 Flash Live, a new voice and audio AI model positioned as the company's best-performing audio offering to date. The model is now available through the Gemini Live API, Google AI Studio, Gemini Live, and Search Live across over 200 countries.
Performance and Capabilities
According to Artificial Analysis benchmarking, Gemini 3.1 Flash Live achieves 95.9% on the Big Bench Audio Benchmark when configured to its "High" thinking level, placing it second only to Step-Audio R1.1 Realtime, which scores 97.0%. At the "Minimal" thinking level, the model's score drops to 70.5% but response time improves significantly.
Response latency varies by configuration:
- High thinking: 2.98-second response time
- Minimal thinking: 0.96-second response time
Google claims the model delivers improved pitch and emotion detection compared to its predecessor, with enhanced reliability in noisy environments. Developers can now configure thinking levels directly, allowing trade-offs between output quality and latency.
Pricing
Gemini 3.1 Flash Live maintains identical pricing to its Gemini 2.5 predecessor:
- Audio input: $0.35 per hour
- Audio output: $1.40 per hour
Google positions this as among the cheapest audio AI models available. For comparison, Step-Audio R1.1 Realtime offers lower input pricing but charges more for audio output.
Deployment and Integration
The model now powers live mode functionality within the Gemini app, enabling real-time voice conversations. Integration is available through multiple access points, supporting developers building voice applications across the Google ecosystem.
What this means
Gemini 3.1 Flash Live competes directly with Step-Audio R1.1 Realtime in the high-performance voice AI space, with nearly matching benchmark scores at a lower price point. The configurable thinking levels provide developers genuine flexibility for latency-sensitive applications—a meaningful improvement over fixed-performance models. At 0.96 seconds for minimal thinking, the model targets real-time conversational use cases where sub-second response times matter. The widespread availability across 200+ countries and multiple access methods signals Google's commitment to voice as a core interaction paradigm for Gemini products.
Related Articles
Nous Research releases Arete-27B, a Qwen3.8-27B fine-tune that cuts propaganda rate from 22% to 0.9%
Nous Research fine-tuned Alibaba's Qwen3.8-27B on about 53 million tokens to remove Chinese state propaganda from its answers. According to Nous, the resulting Arete-27B model drops the propaganda rate from 22 percent to 0.9 percent while keeping general capabilities intact. Weights, test set and training data are public under Apache 2.0.
PrunaAI releases P-H3-Image: 8 MP image generation and editing in 8 steps, built on MiniMax-H3
PrunaAI has released P-H3-Image, an image model based on MiniMax-H3 that renders 8-megapixel images in 8 steps with no classifier-free guidance. It handles text-to-image and editing with up to 8 input images, and bf16 inference needs about 101 GB of GPU memory.
OpenCSG Agentic-27B: 27.36B-Parameter Tool-Use Model Runs on One A800, Commercial Use Needs Permission
OpenCSG has released Agentic-27B, a 27.36B-parameter model for multi-step tool use, built on the dense Qwen3.8-27B. The company reports it runs on a single 80GB NVIDIA A800, but commercial use requires authorization under the OpenCSG license, and all benchmark results are OpenCSG's own.
Salesforce launches Koa Decisions, a Gemma-based model that classifies text, image and audio requests
Salesforce introduced Koa Decisions on November 6, 2026. The decision model is built from Google's open-weight Gemma and accepts text, images and audio. It is the next version of HyperClassifier.
Comments
Loading...