3 min remaining
0%
人工智慧與機器學習

零的競賽:小米剛剛投下了一顆AI核彈

小米對其新AI模型的激進定價可能重新定義市場,提供95%到99%的折扣,挑戰AI經濟的現狀。

3 min read
Progress tracked
3 分鐘閱讀·

簡而言之:當西方專注於模型能力(智商)時,小米卻改變了話題,轉向單位經濟學。通過將他們的新模型定價為$0.10每百萬標記——這比市場標準便宜驚人的20倍到100倍——他們不僅僅是在競爭;他們試圖完全商品化智慧。這讓人想起14年前他們的智慧型手機策略。問題不是「這個模型好嗎?」而是:「誰能在這些價格下生存?」

我是James,Mercury Technology Solutions的首席執行官。

在武術界,有一句話:「不講武德」。這意味著不光明正大地打鬥——在對手倒下時踢他,或使用壓倒性的力量瞬間結束戰鬥。

小米剛剛進入AI聊天,完全沒有武德。

不可能的數學

讓我們看看小米剛剛在板上寫下的數字:

  • 價格: $0.10美元 / 每百萬標記
  • 速度:150標記/秒

作為參考,目前市場上頂級的封閉源模型通常起價為$2.00到$10.00每百萬標記。小米並不是在提供20%的折扣。他們提供的是95%到99%的折扣。

"便宜"不等於"不好"

通常,當某樣東西便宜100倍時,它是垃圾。但MiMo-V2-Flash並不是玩具。它在SWE-bench(軟體工程基準)上得分為73.4%。這使它距離OpenAI和Anthropic的頂級封閉模型只有一步之遙。它不是房間裡最聰明的,但對90%的商業用例來說,它「足夠聰明」——而且運行成本幾乎為零。

小米的遊戲手冊:2011年重演

如果你在科技界待得夠久,這感覺像是似曾相識。

14年前,小米以

1,999人民幣推出了他們的第一款智慧型手機。當時,可比的旗艦機型價格是其兩倍或三倍。他們將硬體商品化,摧毀了競爭對手的利潤,並在生態系統上建立了一個帝國。現在,他們正對

標記經濟做同樣的事情。他們打賭如果將智慧的成本壓到零,他們將擁有下一個十年的基礎設施。正如小米的羅福立所說:

「這只是AGI路線圖的第二步。」誰能在價格戰中生存?

這對AI產業來說是一個可怕的信號。

中層包裝商:

  1. 如果你的商業模式是以稍微加價的方式轉售GPT-4的訪問權限,那你就死定了。你無法與0.10美元競爭。專有模型建設者:
  2. 除非你擁有Google或Meta的規模,否則當你的競爭對手以虧損出售推理時,你如何證明燒錢訓練模型的合理性?小米已經宣告了「高利潤AI」時代的結束。我們正在進入

實用時代。智慧正變得像電一樣便宜。. Intelligence is becoming as cheap as electricity.

贏家不會是那些銷售電力的人。贏家將是那些建造依賴電力運行的工廠的人。

水銀科技解決方案:加速數位化。

Frequently Asked Questions

What pricing strategy is Xiaomi using for its new AI model?

Xiaomi has adopted an aggressive pricing strategy, offering its new AI model at just $0.10 per million tokens. This pricing is significantly lower than the market standard, which typically ranges from $2.00 to $10.00 per million tokens, representing a staggering 95% to 99% discount.

How does Xiaomi's AI model compare to existing competitors?

Despite its low cost, Xiaomi's MiMo-V2-Flash model scores 73.4% on the Software Engineering Benchmark (SWE-bench), positioning it competitively against top-tier models from companies like OpenAI and Anthropic. While it may not be the absolute best, it is deemed 'smart enough' for 90% of commercial use cases.

What does this pricing shift mean for the AI industry?

Xiaomi's pricing strategy signals a significant shift in the AI industry, moving towards a 'Utility Era' where intelligence is commoditized, much like electricity. Companies that rely on high-margin AI models may struggle to survive as the cost of intelligence decreases to nearly zero.

What implications does this have for businesses using AI?

Businesses that currently operate by reselling AI access or developing proprietary models without the scale of giants like Google or Meta will find it increasingly difficult to compete. The competitive landscape will favor those who can build infrastructure around this low-cost intelligence rather than those selling it at a premium.

How does Xiaomi's approach mirror its smartphone strategy?

Xiaomi's current strategy echoes its initial smartphone launch 14 years ago, where it offered flagship devices at dramatically lower prices than competitors. By commoditizing AI token economics in a similar fashion, Xiaomi aims to dominate the infrastructure of the future, just as it did in the smartphone market.