Plain-English Summary
AWS’s speech-to-speech model for natural, low-latency voice conversations.
Amazon Nova 2 Sonic
AWS’s speech-to-speech model for natural, low-latency voice conversations.
Nova 2
Audio / Speech
GA
Open weights: No
API: Yes — Amazon Bedrock
Technical Notes
Streaming model that accepts speech and text and produces speech and text for real-time conversational applications.
Best For
Voice agents, call centers, conversational assistants
Watch Out For
AWS-specific integration; voice-agent quality depends on latency, telephony, and interruption handling.
Model Specs
- Maker
- Amazon Web Services
- Country
- Not listed
- Family
- Nova 2
- Type
- Audio / Speech
- Release date
- 2025-12-02
- Date confidence
- Exact
- Status
- GA
- Context window
- Varies
- Max output
- Not listed
- Parameters
- Not disclosed
- Active parameters
- Not disclosed
- Architecture
- Proprietary streaming speech model
- Reasoning
- Conversation planning
- Tool calling
- Yes / agent integration
- Structured output
- Endpoint dependent
- API available
- Yes — Amazon Bedrock
- Open weights
- No
- Self-hostable
- No
- License
- Proprietary
- Inputs
- Speech, text
- Outputs
- Speech, text
- Approx. price
- See provider pricing
- How to access
- Provider website
- Model / API ID
- amazon.nova-2-sonic
- Knowledge cutoff
- Not publicly disclosed
- Fine-tuning
- Voice-agent configuration, tools, RAG
- Completeness
- 81 %
- Verified
- Jul 18, 2026
- Snapshot date
- Jul 18, 2026
- Source
- Official source