Nova 2 Lite
Language Available Full comparison ↗ amazon/nova-2-lite · by Amazon
· decoder-only-transformer
Pricing — 1 offering(s)
Input tokens
- $0.30 / 1M tokens (input)
Current
2025-12-02 → present
AWS's own pricing page renders per-model rates via a client-side JS widget (`{priceOf!...}` tokens) rather…
AWS's own pricing page renders per-model rates via a client-side JS widget (`{priceOf!...}` tokens) rather than static HTML — confirmed by fetching the raw page source directly, which shows the Nova 2 Lite row present but with unresolved price tokens rather than literal numbers. This figure is cross-verified against three independent third-party cost trackers that scrape Bedrock's live pricing (avahi.ai, futureagi.com llm-cost-calculator, pricepertoken.com), all of which agree on $0.30/$2.50 per 1M input/output tokens.
Output tokens
- $2.50 / 1M tokens (output)
Current
2025-12-02 → present
Same cross-verification caveat as the input tier — see its notes.
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Nova 2 Lite fits into a cost-aware routing setup
See how →Capability profile
How Nova 2 Lite rates across core capability dimensions, with the task-level evidence behind each rating.
AWS's own framing is "a fast, cost-effective reasoning model." Amazon's Nova 2 report puts Nova 2 Lite around 61.8 on MMMU-Pro (visual reasoning) — ahead of Claude Haiku, behind GPT-5 mini and Gemini Flash. The adjustable-thinking control lets it stretch further on hard tasks at higher cost.
Function calling / tools supported via Bedrock; positioned for document processing and business automation agents.
1,000,000-token context window (Amazon Nova 2 materials / community spec trackers).
The speed/cost tier of the Nova 2 line — designed for high-volume everyday workloads.
Operator guidance
The default Nova 2 choice for cost- and latency-sensitive production workloads on AWS, especially multimodal or very-long-context ones. Step up to Nova 2 Pro (Preview on Bedrock at time of writing) for harder reasoning. Outside AWS, Gemini 2.5 Flash and GPT-5.x mini are the closest cheap-multimodal-long-context alternatives and score somewhat higher on visual reasoning; Nova 2 Lite's edge is price and native Bedrock/AWS integration.
Use cases
- High-volume customer-service chatbots, document processing, and business-process automation where cost and latency matter more than frontier quality
- Multimodal workloads (text + image + video input) at low cost within an AWS/Bedrock stack
- Long-document / large-corpus tasks needing up to 1M tokens of context
- Agentic workflows where the adjustable thinking budget can be tuned per task
Limitations
- Bedrock / AWS-hosted only — no independent first-party API outside AWS
- Visual-reasoning quality (MMMU-Pro ~61.8) trails GPT-5 mini and Gemini Flash in Amazon's own comparison
- Pricing is not on static HTML on AWS's pricing page (client-rendered) — see registry entry for the Price List API cross-check
- Capability ratings here are qualitative, drawn from Amazon's own technical report and framing; no independent third-party leaderboard entry consulted for this doc