--- title: US companies are routing routine AI tasks to DeepSeek... description: A significant price gap is emerging in the AI market, with US companies like Coinbase, DoorDash, Airbnb, and Cursor adopting 'thrift-maxxing' strategies —... canonical: https://daily.dev/posts/us-companies-are-routing-routine-ai-tasks-to-deepseek-and-other-cheap-chinese-models-i9jznoaxf twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: US companies are routing routine AI tasks to DeepSeek and other cheap Chinese models | daily.dev og:description: A significant price gap is emerging in the AI market, with US companies like Coinbase, DoorDash, Airbnb, and Cursor adopting 'thrift-maxxing' strategies —... og:url: https://daily.dev/posts/us-companies-are-routing-routine-ai-tasks-to-deepseek-and-other-cheap-chinese-models-i9jznoaxf og:image: https://api.daily.dev/og/posts/I9jZnOaXf.png og:image:alt: US companies are routing routine AI tasks to DeepSeek and other cheap Chinese models og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # US companies are routing routine AI tasks to DeepSeek and other cheap Chinese models **[Collections](https://daily.dev/sources/collections)** · 3 min read · 1 upvotes · 2 comments ## Summary A significant price gap is emerging in the AI market, with US companies like Coinbase, DoorDash, Airbnb, and Cursor adopting 'thrift-maxxing' strategies — routing routine AI tasks to cheap Chinese models like DeepSeek (at ~$0.87 per 750K words) while reserving expensive frontier models for critical work. In one week this July, Chinese models captured 57% of tokens consumed by US firms on OpenRouter. This trend threatens the premium pricing models that underpin OpenAI and Anthropic's IPO valuations. Friction exists via cloud providers (AWS, Azure, Google) not yet supporting Chinese open-weight models and Congressional scrutiny, but a 57x price difference is hard to ignore for most engineering teams. ## Content ## What dropped DeepSeek released V4-Flash-0731, a weight update to their efficiency-oriented Flash variant, and put the official API into public beta. The architecture is a sparse mixture-of-experts with 256 routed experts, 6 active per token, 284B total parameters, 13B active, and a 1M-token context window. It supports three reasoning-effort levels and was built specifically for code agents and tool use. For anyone already running the DSpark preview checkpoint, the serving config carries over unchanged. The DSpark draft module is bundled inside the weights, so speculative decoding is a single flag rather than a second model to load: ``` --speculative-config '{"method":"dspark","num_speculative_tokens":7,"draft_sample_method":"greedy"}' ``` ## The numbers that matter Pricing: $0.14 per million input tokens, $0.28 per million output tokens. For context, 750,000 words of AI output costs roughly $50 from Anthropic and about $0.87 from DeepSeek. On TerminalBench-2.1, this checkpoint jumped more than 20 points over the previous Flash preview. On the AA Index it's about 10 points ahead of the prior checkpoint, putting it roughly level with GLM 5.2 and GPT Luna on capability benchmarks while being substantially cheaper than either. On Hugging Face's inference endpoints it runs at around 400 tokens per second for $10/hour. Some versions can run locally on a 96GB Mac Studio or a DGX Spark. That's $4,000 hardware running something that benchmarks near frontier. ## The price war context This release didn't happen in a vacuum. OpenAI cut GPT-5.6 Luna pricing around the same time, and the broader pattern is hard to miss: Chinese models captured 57% of tokens consumed by US firms on OpenRouter in one July week. Companies like Coinbase, DoorDash, Airbnb, and Cursor are running "orchestration" strategies that route routine tasks to cheap models and reserve expensive ones for complex work. This is putting real pressure on OpenAI and Anthropic's valuations ahead of potential IPOs. Both have responded with free token giveaways and cheaper model releases, but the gap is still wide. ## The risks people aren't ignoring Sequoia's Dean Meyerrr laid out two concerns worth taking seriously. The first is dependency. If your product's capability roadmap depends on a Chinese lab releasing the next model, you're not in control of your own trajectory. That's a real strategic problem regardless of how good the current model is. The second is security. Verifying whether a model contains a backdoor is, as he put it, an NP-hard problem. You can't audit your way to certainty. He compared the current state of AI security to early internet days when Wi-Fi encryption was publicly broken and nobody acted on it. Major cloud providers - AWS, Azure, Google - don't yet support Chinese open-weight models, and Congress is actively probing companies that use them. None of that makes the price argument go away. It just means the tradeoffs are real on both sides. ## Where to run it The official API is live in public beta directly from DeepSeek. ZenMux is offering it free for a limited time and claims to be the only provider running the official 0731 version. vLLM supports it with the architecture carrying over from the DSpark preview. ## Community discussion Top comments from developers on daily.dev. **@petecapecod** · 1 upvotes > yeah security is a factor, but with these open weight models you can spin them up on US infra which rules out any telemetry. Because you would see the outbound calls. > > That's what Cursor does with theirs. **@devfredx** · 0 upvotes > Honestly speaking whatever problem this article talked about that is a risk factor for using open weight is still an issue in closed models. Oh and also stop using security as an excuse to dismiss open weight models when open AI closed models is busy going rogue. I'm tired of this nonsense --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#openai](https://daily.dev/tags/openai), [#anthropic](https://daily.dev/tags/anthropic), [#deepseek](https://daily.dev/tags/deepseek) [View this post on daily.dev](https://daily.dev/posts/us-companies-are-routing-routine-ai-tasks-to-deepseek-and-other-cheap-chinese-models-i9jznoaxf) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"US companies are routing routine AI tasks to DeepSeek and other cheap Chinese models","url":"https://daily.dev/posts/us-companies-are-routing-routine-ai-tasks-to-deepseek-and-other-cheap-chinese-models-i9jznoaxf","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/us-companies-are-routing-routine-ai-tasks-to-deepseek-and-other-cheap-chinese-models-i9jznoaxf"},"datePublished":"2026-07-27T17:25:09.563Z","dateModified":"2026-07-31T19:03:38.908Z","description":"A significant price gap is emerging in the AI market, with US companies like Coinbase, DoorDash, Airbnb, and Cursor adopting 'thrift-maxxing' strategies —...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":2,"discussionUrl":"https://daily.dev/posts/us-companies-are-routing-routine-ai-tasks-to-deepseek-and-other-cheap-chinese-models-i9jznoaxf","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":2}],"keywords":"ai-agents,openai,anthropic,deepseek","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"US companies are routing routine AI tasks to DeepSeek and other cheap Chinese models"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/us-companies-are-routing-routine-ai-tasks-to-deepseek-and-other-cheap-chinese-models-i9jznoaxf","comment":[{"@type":"Comment","text":"yeah security is a factor, but with these open weight models you can spin them up on US infra which rules out any telemetry. Because you would see the outbound calls.\nThat’s what Cursor does with theirs.","datePublished":"2026-07-28T12:34:11.405Z","url":"https://daily.dev/posts/I9jZnOaXf#c-Jfx8bELb9","author":{"@type":"Person","name":"Peter Cruckshank","url":"https://daily.dev/petecapecod","image":"https://media.daily.dev/image/upload/s--ZJhQyKws--/f_auto/v1721235024/avatars/avatar_A9xh33q0QoxtkGoJRCosp"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}},{"@type":"Comment","text":"Honestly speaking whatever problem this article talked about that is a risk factor for using open weight is still an issue in closed models. Oh and also stop using security as an excuse to dismiss open weight models when open AI closed models is busy going rogue. I’m tired of this nonsense","datePublished":"2026-07-29T08:13:48.768Z","dateModified":"2026-07-29T08:14:12.514Z","url":"https://daily.dev/posts/I9jZnOaXf#c-lkgPeEban","author":{"@type":"Person","name":"Fredrick-Marie","url":"https://daily.dev/devfredx","image":"https://avatars.githubusercontent.com/u/78586515?v=4"}}]} ```