Amazon Nova 2 Lite

Nova 2 Lite is a capable cost-oriented multimodal reasoning model rather than a frontier general-purpose leader. Amazon reports multimodal input, optional extended thinking, and a 1M-token context window; Artificial Analysis lists 145.5 output tokens/s, 1.23s TTFT, and $0.30/$2.50 per million input/output tokens. This supports stronger cost and speed scores than Grok 3, while its documented Bedrock integration supports a developer experience near Kimi K2.5. Coding evidence is material but qualified: Amazon’s report gives 53.6% SWE-Bench Verified under a standard setup and 64.5% with inference-time scaling plus internal agent scaffolding, alongside 71.0 on a specified LiveCodeBench v5 subset. That places it above low-evidence or weaker coding options such as Hermes 4.3 36B, but below Kimi K2.5, Grok 3, GPT-5.5, and Claude Opus 4.8, which have substantially higher calibrated coding/agentic scores. Its broad multimodal and long-context positioning is credible, but does not match Gemini 3.5 Flash’s stronger multimodal anchor. Pricing clarity is limited because the supplied official material does not provide a price table; the price figures come from Artificial Analysis. Public adoption and preference evidence is also immature: neither DeepSWE nor the checked Arena-Hard leaderboard lists Nova 2 Lite, and no Terminal-Bench/Aider result was supplied. Scores therefore credit official documentation, Amazon-reported benchmark detail, and independent performance/price tracking, while discounting vendor-reported results and unavailable external replications.