8月6日，两条大模型行业的新闻放在一起看，特别有戏剧性。
一条来自OpenAI：ChatGPT免费用户的默认模型正式升级为GPT-5.6 Luna，而且Chat模式不再有使用次数限制——想聊多久聊多久，想聊多少轮聊多少轮，完全免费。Plus和Pro用户的5.6 Sol也同步获得专精升级，事实准确性更高、回答更聚焦，还新增了思考深度调节滑块。
另一条来自DeepSeek：公司在开发者后台发布公告，称计划近期整体上调DeepSeek API服务的定价，预计涨幅较大，请开发者合理安排使用。没有说涨多少，但"涨幅较大"这个措辞已经够让人紧张了。
一边是全球最大的大模型公司用顶级模型免费送的方式抢用户，一边是中国最有竞争力的独立大模型公司扛不住成本要涨价。大模型行业的冰火两重天，从来没有像今天这么清晰过。
免费的底气：巨头的补贴逻辑
OpenAI为什么敢把GPT-5.6 Luna免费放出来？因为它的商业模式和独立玩家完全不一样。
对于OpenAI来说，免费版ChatGPT不是一个赚钱的产品，而是一个获客漏斗和数据金矿。免费用户用得越多，产生的对话数据越多，模型就能迭代得越好；免费用户里转化为Plus/Pro/Enterprise付费用户的比例越高，整体ROI就越好看。而且ChatGPT有超过两亿月活用户，这个流量入口本身就有巨大的商业价值——未来不管是做搜索、做应用商店、做企业服务，都是基于这个用户池。
更关键的是，OpenAI有微软这个爸爸在背后撑着。算力成本、数据中心建设、研发投入，这些都是百亿美元级别的开支，没有几个公司能扛得住长期补贴。但微软可以——它把OpenAI的AI能力整合进Azure、Office、Windows、Bing，从别的地方把钱赚回来。对微软来说，ChatGPT免费不是亏损，是战略投资。
而DeepSeek的处境完全不同。作为一家独立的大模型公司，它没有微软那样的生态互补，也没有两亿C端用户做漏斗。它的主要收入来源就是API调用费。当行业竞争加剧、算力成本居高不下、价格战把利润空间越打越薄的时候，涨价几乎是必然的选择。
这不是第一家涨价的中国大模型公司。上海证券报的数据显示，智谱AI 2026年一季度API价格已经上涨了83%。DeepSeek这次跟进，说明中国大模型行业的低价倾销阶段已经走到头了。
价格分化背后：行业加速洗牌
大模型行业正在经历一场残酷的分化。
一端是有生态、有流量、有钱的巨头——OpenAI+微软、Google、Meta、亚马逊。它们可以用补贴策略快速抢占市场份额，把模型能力下放到免费用户，用规模效应摊薄成本。它们不靠卖API赚钱，它们靠生态赚钱。
另一端是独立大模型公司——Anthropic、Cohere、DeepSeek、智谱、月之暗面，等等。它们的核心资产是模型能力，核心商业模式是卖API。但当巨头把模型能力免费送的时候，独立公司的生存空间就被急剧压缩了。
出路在哪里？大概有三条。
第一，做垂直、做行业、做巨头不愿做的脏活累活。通用大模型是巨头的战场，但法律、医疗、金融、工业这些垂直领域，巨头很难深耕。谁能在某个垂直领域做到不可替代，谁就能活下来。
第二，做Agent、做应用、从卖模型变成卖解决方案。单纯卖token的生意越来越难做了，但如果能把模型包装成具体的产品和服务，价值就能往上走。这也是为什么现在所有大模型公司都在往Agent方向转。
第三，拥抱巨头、被收购、或者成为巨头生态的一部分。这可能是大多数独立大模型公司最终的归宿——不是每家公司都能成为OpenAI，但做OpenAI生态里的重要玩家，也不失为一个好的结局。
DeepSeek的涨价不是终点，只是开始。未来一两年，大模型行业的洗牌会越来越剧烈。能活下来的，要么是有生态的巨头，要么是有绝活的垂直玩家。中间那些不上不下的，会很难受。
明天见。
August 6. Two LLM industry headlines, read side by side, are strikingly dramatic.
One comes from OpenAI: ChatGPT's free tier is officially upgrading its default model to GPT-5.6 Luna, and Chat mode will no longer have usage limits — chat as long as you want, as many turns as you want, completely free. Plus and Pro users also get a specialized upgrade to 5.6 Sol with better factual accuracy, more focused answers, and a new thinking-depth slider.
The other comes from DeepSeek: the company posted an announcement in its developer portal saying it plans to overall raise DeepSeek API service pricing in the near term, with a relatively large increase expected, advising developers to plan their usage accordingly. They didn't say how much, but the phrase "relatively large increase" is enough to make people nervous.
On one side, the world's largest LLM company is giving away its top-tier model for free to grab users. On the other side, China's most competitive independent LLM company can't sustain costs and has to raise prices. The LLM industry's tale of two realities has never been clearer than it is today.
The Confidence Behind Free: The Giant's Subsidy Logic
Why does OpenAI dare to release GPT-5.6 Luna for free? Because its business model is completely different from independent players.
For OpenAI, free ChatGPT isn't a revenue-generating product — it's a customer acquisition funnel and a data goldmine. The more free users use it, the more conversation data gets generated, the better the model iterates. The higher the conversion rate from free to Plus/Pro/Enterprise paid users, the better the overall ROI. And with over 200 million monthly active users, ChatGPT itself has enormous commercial value as a traffic entry point — whether for search, app stores, or enterprise services down the line, everything builds on this user base.
More critically, OpenAI has Microsoft backing it. Compute costs, data center construction, R&D investment — these are all hundred-billion-dollar expenses, and few companies can sustain long-term subsidies. But Microsoft can — it integrates OpenAI's AI capabilities into Azure, Office, Windows, and Bing, and makes the money back elsewhere. For Microsoft, free ChatGPT isn't a loss; it's strategic investment.
DeepSeek's situation is entirely different. As an independent LLM company, it doesn't have Microsoft's ecosystem complementarity, nor does it have 200 million consumers as a funnel. Its main revenue source is API call fees. When industry competition intensifies, compute costs stay high, and price wars squeeze profit margins thinner and thinner, price hikes are almost inevitable.
This isn't the first Chinese LLM company to raise prices. Shanghai Securities News data shows that Zhipu AI's API prices already rose 83% in Q1 2026. DeepSeek following suit confirms that the Chinese LLM industry's low-price dumping phase has reached its end.
Behind Price Stratification: Industry Shuffling Accelerates
The LLM industry is undergoing a brutal stratification.
On one end are the ecosystem, traffic, and capital giants — OpenAI+Microsoft, Google, Meta, Amazon. They can use subsidy strategies to rapidly capture market share, push model capabilities down to free users, and amortize costs through scale effects. They don't make money selling APIs; they make money from their ecosystems.
On the other end are independent LLM companies — Anthropic, Cohere, DeepSeek, Zhipu, Moonshot, and others. Their core asset is model capability, their core business model is selling APIs. But when giants give away model capability for free, independent companies' living space gets dramatically compressed.
What's the way out? Roughly three paths.
First, go vertical, go industry, do the messy, hard work that giants won't touch. General-purpose LLMs are the giants' battlefield, but in vertical domains like law, healthcare, finance, and industry, giants struggle to go deep. Whoever becomes irreplaceable in a vertical domain survives.
Second, build agents, build applications, shift from selling models to selling solutions. The pure token-selling business is getting harder and harder, but if you can package models into concrete products and services, value moves upmarket. This is why every LLM company is pivoting toward agents these days.
Third, embrace the giants, get acquired, or become part of a giant's ecosystem. This might be the final destination for most independent LLM companies — not every company can become OpenAI, but being an important player in the OpenAI ecosystem isn't a bad outcome either.
DeepSeek's price hike isn't the end — it's just the beginning. Over the next year or two, the LLM industry shakeout will only intensify. The survivors will either be ecosystem giants or vertical specialists with unique strengths. Everyone stuck in the middle will have a hard time.
See you tomorrow.
大模型商业 · ChatGPT · GPT-5.6 Luna · OpenAI · DeepSeek · API涨价 · 价格战 · 行业洗牌 · 免费策略 · 补贴逻辑
LLM Business · ChatGPT · GPT-5.6 Luna · OpenAI · DeepSeek · API Price Hike · Price War · Industry Shakeout · Free Strategy · Subsidy Logic
