Amazon on Wednesday released what the company claims is the most capable AI model in its Nova family, Nova Premier.
Amazon 在禮拜三推出了他們家 Nova 系列裡，號稱目前最強的 AI 模型，叫做 Nova Premier。
Nova Premier, which can process text, images, and videos (but not audio), is available in Amazon Bedrock, the company’s AI model development platform. Amazon says that Premier excels at “complex tasks” that “require deep understanding of context, multi-step planning, and precise execution across multiple tools and data sources.”
這個 Nova Premier 可以處理文字、圖片跟影片（但不能處理聲音），已經可以在 Amazon Bedrock 上用了。Bedrock 是他們家開發 AI 模型的平台。Amazon 說，Premier 超會處理那種「很複雜的任務」，就是需要「深入理解上下文、多步驟規劃，還要很精準地串接各種工具跟資料來源」的那種。
Amazon announced its Nova lineup of models in December at its annual AWS re:Invent conference. Over the last few months, the company has expanded the collection with image- and video-generating models as well as with audio understanding and agentic, task-performing releases.
Amazon 是去年 12 月在他們一年一度的 AWS re:Invent 大會上發表了 Nova 系列模型。這幾個月來，他們又陸續推出可以生成圖片、影片的模型，還有能理解聲音、執行任務的 agent 式模型，把這個系列擴展得更完整。
Nova Premier, which has a context length of 1 million tokens, meaning it can analyze around 750,000 words in one go, is weaker on certain benchmarks than flagship models from rival AI companies such as Google. On SWE-Bench Verified, a coding test, Premier is behind Google’s Gemini 2.5 Pro, and it also performs poorly on benchmarks measuring math and science knowledge, GPQA Diamond and AIME 2025.
Nova Premier 的上下文長度是 100 萬 tokens，換句話說，它可以一次讀懂大約 75 萬個字。不過，跟 Google 這種競爭對手家的旗艦模型比起來，Premier 在某些測試上就沒那麼威。像是 SWE-Bench Verified 這個寫程式的測試，Premier 就輸給 Google 的 Gemini 2.5 Pro；在測數學和科學知識的 GPQA Diamond 和 AIME 2025 上，表現也比較弱。
However, in bright spots for Premier, the model does well on tests for knowledge retrieval and visual understanding, SimpleQA and MMMU, according to Amazon’s internal benchmarking.
不過，Premier 也是有亮點啦！根據 Amazon 自己內部測試的結果，這個模型在知識檢索和視覺理解方面的測試，像是 SimpleQA 和 MMMU，表現都還不錯。
In Bedrock, Premier is priced at $2.50 per 1 million tokens fed into the model and $12.50 per 1 million tokens generated by the model. That’s around the same price as Gemini 2.5 Pro, which costs $2.50 per million input tokens and $15 per million output tokens.
在 Bedrock 上用 Premier，輸入 100 萬 tokens 算 $2.50 美元，輸出 100 萬 tokens 算 $12.50 美元。這個價錢跟 Gemini 2.5 Pro 差不多，Gemini 2.5 Pro 輸入 100 萬 tokens 是 $2.50 美元，輸出 100 萬 tokens 是 $15 美元。
Importantly, Premier isn’t a “reasoning” model. As opposed to models like OpenAI’s o4-mini and DeepSeek’s R1, it can’t take additional time and computing to carefully consider and fact-check its answers to questions.
有個重點是，Premier 不是那種「推理」模型。它不像 OpenAI 的 o4-mini 或 DeepSeek 的 R1 那樣，可以花更多時間和運算資源去仔細思考和查證問題的答案。
Exhibit at TechCrunch Sessions: AI
TechCrunch Sessions: AI 的展覽
Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last.
趕快報名參加 TC Sessions: AI，把你做出來的東西秀給 1200 位以上的決策者看，不用花大錢！只到 5 月 9 號，或是桌子沒了就沒了。
Exhibit at TechCrunch Sessions: AI
Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last.
柏克萊，加州 | 6 月 5 日
Amazon is pitching Premier as best for “teaching” smaller models via distillation — in other words, transferring its capabilities for a specific use case into a faster, more efficient package.
Amazon 主打 Premier 模型最適合拿來「教」小模型，透過蒸餾（distillation）的方式，把它的厲害之處轉移到針對特定用途、更快、更有效率的套件裡面。
Amazon sees AI as increasingly core to its overall growth strategy. CEO Andy Jassy recently said the company is building more than 1,000 generative AI applications and that Amazon’s AI revenue is growing at “triple-digit” year-over-year percentages and represents a “multi-billion-dollar annual revenue run rate.”
Amazon 越來越覺得 AI 是他們整體成長策略的核心。CEO Andy Jassy 最近說，公司正在開發超過 1,000 個生成式 AI 應用程式，而且 Amazon 的 AI 營收年成長率高達「三位數」，每年營收「上看數十億美元」。
