Generational successor to the Nova 1 family, launched at AWS re:Invent 2025 (December 2). Adds extended thinking with three configurable budgets (low / medium / high), a 1M-token context window across the reasoning tiers, and built-in web grounding plus a code interpreter. Authored by the Amazon AGI org under Peter DeSantis (heading the unified AGI/Silicon/Quantum group from December 2025).

The family covers four models. Nova 2 Pro (preview only) is the most capable reasoner. Nova 2 Lite is the cost-tier reasoner at the same context length. Nova 2 Omni is a unified multimodal reasoner that accepts text, images, video, and audio as input and produces both text and image outputs — the first Nova entry with native image generation inside a reasoning loop. Nova 2 Sonic is the speech-to-speech model (its own page).

Top AA Intelligence Index v4.3 score in the family is 14 (Nova 2 Pro Preview, medium thinking) — a meaningful jump over the Nova Premier predecessor (9). Proprietary, available via Amazon Bedrock.

Model Details

Context window 1,000,000
AA Intelligence 14 on v4.3.2 was 16 on v4.2

Variants

Name Parameters Notes
Nova 2 Pro Preview (medium thinking) — AA Intelligence Index v4.3: 14 (top of family); 1M context; reasoning + web grounding + code interpreter
Nova 2 Pro Preview (low thinking) — AA Intelligence Index v4.3: 13
Nova 2 Pro Preview (non-reasoning) — AA Intelligence Index v4.3: 10; 256K context for non-reasoning mode
Nova 2 Lite (high thinking) — AA Intelligence Index v4.3: 13; 1M context; cost-tier reasoner
Nova 2 Lite (medium thinking) — AA Intelligence Index v4.3: 12
Nova 2 Lite (low thinking) — AA Intelligence Index v4.3: 12
Nova 2 Lite (non-reasoning) — AA Intelligence Index v4.3: 9
Nova 2 Omni (medium thinking) — AA Intelligence Index v4.3: 14; native multimodal in/out (text+image+video+audio → text+image)
Nova 2 Omni (low thinking) — AA Intelligence Index v4.3: 11
Nova 2 Omni (non-reasoning) — AA Intelligence Index v4.3: 8
frontiermultimodalreasoningagents

Related