DeepSeek V4's Price Cut Makes Frontier AI Nearly Free

DeepSeek V4 just reset the floor on AI pricing. In 2026 the company made its V4 price cut permanent, dropping API costs by roughly 75% to about $0.27 per million input tokens — the cheapest frontier-grade model on the market, and still completely free to use on the web. Here's what changed and what it means for you.

What changed

The headline is simple: the temporary discount DeepSeek floated earlier is now the standing rate. With V4, the open-weight model with near-frontier reasoning and coding, API pricing lands at roughly $0.27 per million input tokens and $1.10 per million output tokens. That's about a 75% cut from where comparable access used to sit.

If you don't touch the API at all, it's even simpler — the web app and mobile app are free, with no visible query cap. You get V4's full reasoning, a 1M-token context window (around 750,000 words), and unlimited everyday chats without paying a cent. The open weights remain free to download and self-host, too, so there are now three ways to run V4 and only one of them costs money.

Why it matters

Cheap chatbots aren't new. Frontier-grade reasoning at this price is. Most flagship consumer plans converged around $20/month in 2026, and API rates from the big labs are priced for businesses, not tinkerers. DeepSeek undercuts all of them while scoring in the same league on reasoning and coding benchmarks — and it hands you a million-token context for free, something rivals reserve for paid tiers.

That puts real pressure on API pricing across the industry. When a credible, frontier-class model costs a fraction of what OpenAI, Anthropic and Google charge per token, the incumbents have to justify the gap on ecosystem, reliability and trust rather than raw capability. For anyone building on top of an LLM, the math just shifted: workloads that were too expensive to run at scale are suddenly viable. For a sense of where it sits against the field, see our best AI chatbots guide.

The catch

The price is the easy part. The trade-offs are the reason DeepSeek isn't an automatic yes.

First, it's Chinese-hosted. Anything you send through the app or the API sits under Chinese jurisdiction, which is a non-starter for a lot of confidential, regulated or enterprise data. Second, the model censors politically sensitive topics — ask about certain subjects and you'll get a deflection or a refusal rather than an answer. Third, the ecosystem is thin: there's no native voice, image generation, plugins or agents, and the hosted service can slow down under heavy load.

The escape hatch is the open weights. Because V4 is open, privacy-sensitive users and teams can self-host it and keep data entirely in their own environment — you lose the convenience of the hosted app but sidestep the jurisdiction problem entirely. For most people doing casual or learning work, the free web app is fine; for confidential or regulated material, self-hosting is the only responsible path.

What it means for you

If you're a student, a developer on a budget, or anyone who balks at $20/month, DeepSeek V4 is the standout. You get frontier-class reasoning and coding for free on the web, a huge context window at no cost, and — if you outgrow the app — the lowest API rates anywhere to build on.

If you're an enterprise or handle sensitive data, weigh the jurisdiction question carefully before piping anything important through the hosted service. The capability is there; the hosting location is the catch, and self-hosting the open weights is your way to keep both the savings and your data.

For the full breakdown of strengths, limits and pricing, read our DeepSeek review. And if you're deciding between the cheapest capable model and the most complete one, our DeepSeek vs ChatGPT comparison settles the closest call.