← Back to Voice models
Amazon (Nova)
Amazon Nova 2.5 Sonic
Released Oct 5, 2026
Amazon's latest speech-to-speech model on Bedrock, the successor to Nova 2 Sonic. It improves reasoning, instruction following, tool-calling accuracy and latency, with the same API, voices and languages.
- Price per minute
- —
- Pricing
- Token-priced on Bedrock, same as Nova 2 Sonic per AWS; exact rates not retrievable from aws.amazon.com/bedrock/pricing (rows did not render).
- Inputs
- audio, text
- Tool calling
- Yes
- Languages
- 7
- Voices
- 18
- Big Bench Audio
- —
- Access
- Generally available
- Licence
- Proprietary
Strengths & weaknesses
Strengths
- Asynchronous tool calling: keeps talking while tools run
- 18 voices, including polyglot voices that speak all 7 languages and code-switch
- Same price as Nova 2 Sonic
- 256K context window
Weaknesses
- Only 7 supported languages; AWS advises against using others
- No speech-to-speech translation, and pitch and accent can't be adjusted
- Not yet on the AA leaderboard
Not notable: Amazon isn't a frontier-lab company and the model isn't on AA. The service card calls it the successor to Nova 2 Sonic with 'the same API surface, voices, and language support', and mentions using open-source models to improve tool calling. Languages: English, Spanish, German, French, Italian, Portuguese, Hindi. Regions: us-east-1, us-west-2, eu-north-1, ap-northeast-1. Price not found: AWS says it matches Nova 2 Sonic, but the Bedrock pricing page's Nova speech rows didn't render for the fetcher.
Links
API model strings
- bedrock
amazon.nova-2-5-sonic