{"slug": "china-s-flood-of-cheap-ai-models-is-pricing-smaller-us-labs-out", "title": "China's Flood of Cheap AI Models Is Pricing Smaller US Labs Out", "summary": "Chinese labs are giving away frontier-grade AI models at a fraction of the cost of US rivals, squeezing smaller American AI startups out of the market. DeepSeek's V4-Pro runs a standard coding benchmark for about $1,071 versus $4,811 for Claude Opus 4.7, and Alibaba's Qwen3-235B prices output tokens at $0.10 per million against $25 per million for Claude Opus 4.8, a 250-fold gap. NBC News reported that roughly 80% of US AI startups now build on Chinese open-source models, and six of the world's top 10 AI models by independent leaderboard ranking are Chinese.", "body_md": "*Chinese labs are giving away frontier-grade AI models for a fraction of what OpenAI and Anthropic charge, and the squeeze is emptying out the middle of the US AI market.*\n\nFlo Crivello didn't wait around for Anthropic to notice. The founder of AI agent startup Lindy switched 100% of the company's traffic from Anthropic's Claude models to DeepSeek's V4 in June. He said on X it saved the 25-person startup millions of dollars, while actually improving performance on many tasks. \"It's a matter of survival for the business,\" he said. He'd switch back if Anthropic cut its prices. It hasn't. Anthropic is raising them instead: Claude Sonnet 5 pricing is set to jump 50% starting September 1, according to The Register.\n\n## The price gap\n\nThat gap between what US labs charge and what Chinese labs charge has become the defining fact of the AI industry in 2026. DeepSeek's V4-Pro runs a standard coding benchmark for about $1,071. The same benchmark on Claude Opus 4.7 costs $4,811, roughly 4.5 times more. The numbers only get worse from there. Alibaba's Qwen3-235B prices its output tokens at $0.10 per million against $25 per million for Claude Opus 4.8, a 250-fold gap cited by The Register. Even DeepSeek's newer release, V4 Flash 0731, undercuts by a smaller but still brutal margin: it lands within a single point of OpenAI's GPT-5.6 Luna on benchmarks while costing 40% less per task, and at 284 billion parameters it's small enough to run on a single enterprise server.\n\nAlibaba didn't slow down either. On August 3, the company released Qwen3.8-Max, a 2.4 trillion-parameter model that Bloomberg reported goes head-to-head with the best systems out of Anthropic and OpenAI, at a price that undercuts both considerably.\n\n## Who's getting squeezed\n\nFrontier labs like OpenAI and Anthropic have brand recognition, distribution, and enterprise contracts to fall back on. Smaller and mid-tier American labs don't have that cushion, and they're the ones getting squeezed out first. Cohere, AI21 Labs, and Reka have spent years building proprietary language models for the US market, and now they're competing against Chinese open-weight systems that any developer can download and run for nothing beyond the compute bill. NBC News reported that roughly 80% of US AI startups now build on top of Chinese open-source models rather than pay for a proprietary American one. That's not a rounding error. Cohere CEO Aidan Gomez told the Financial Times that depending on any single provider carries real risk, a comment that reads differently once you notice how many of his potential customers have already made their choice, and made it in China's favor.\n\nOpenAI has responded the way a company with pricing power responds: it's cutting prices to protect share. It slashed GPT-5.6 Luna's token costs by 80% and cut its mid-tier Terra model by 20%, pricing Terra below Anthropic's Claude Sonnet 4.6 on both input and output tokens. That's a frontier lab defending territory. A startup without OpenAI's balance sheet doesn't have that option. When your entire product is the model itself, and a Chinese lab gives away something comparable for a tenth of the price, there's no margin left to cut into.\n\nSix of the world's top 10 AI models by independent leaderboard ranking are now Chinese, spanning DeepSeek, Moonshot, Zhipu's Z.ai, Tencent, Xiaomi, and MiniMax. Moonshot, the company behind the Kimi models, is closing in on a $30 billion valuation and preparing a Hong Kong listing within six months. Z.ai's stock is up more than 1,100% since its own listing, on the strength of GLM-5.2, a model Silicon Valley engineers have openly praised as close to frontier quality at a fraction of frontier prices.\n\nPalantir's chief technology officer told Bloomberg in July that Chinese AI models now pose a genuine economic risk to the United States, not just a strategic one. That's the real shift here. This isn't a story about China catching up on capability, though it has. It's a story about China setting the price, and every US lab that can't match it is running out of room.\n\n**Also read:** [Alibaba-Backed 3D AI Startup Vast Is Weighing a Hong Kong Listing](https://startupfortune.com/alibaba-backed-3d-ai-startup-vast-is-weighing-a-hong-kong-listing/) • [MoonPay Lets ChatGPT and Claude Spend Your Crypto Behind a Passkey](https://startupfortune.com/moonpay-lets-chatgpt-and-claude-spend-your-crypto-behind-a-passkey/) • [Stat Sniper Wants Its AI Betting Picks to Show Their Work](https://startupfortune.com/stat-sniper-wants-its-ai-betting-picks-to-show-their-work/)", "url": "https://wpnews.pro/news/china-s-flood-of-cheap-ai-models-is-pricing-smaller-us-labs-out", "canonical_source": "https://startupfortune.com/chinas-flood-of-cheap-ai-models-is-pricing-smaller-us-labs-out/", "published_at": "2026-08-04 05:14:43+00:00", "updated_at": "2026-08-04 05:24:41.238795+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-policy", "ai-products", "generative-ai", "ai-startups"], "entities": ["DeepSeek", "Anthropic", "OpenAI", "Alibaba", "Lindy", "Flo Crivello", "Cohere", "Moonshot"], "alternates": {"html": "https://wpnews.pro/news/china-s-flood-of-cheap-ai-models-is-pricing-smaller-us-labs-out", "markdown": "https://wpnews.pro/news/china-s-flood-of-cheap-ai-models-is-pricing-smaller-us-labs-out.md", "text": "https://wpnews.pro/news/china-s-flood-of-cheap-ai-models-is-pricing-smaller-us-labs-out.txt", "jsonld": "https://wpnews.pro/news/china-s-flood-of-cheap-ai-models-is-pricing-smaller-us-labs-out.jsonld"}}