1 hour ago
21
DeepSeek’s latest AI model is the cheapest to run among well-known models. The cost gap isn’t marginal.
The numbers that spooked Wall Street
DeepSeek’s R1 model, which launched in January 2025, operates at roughly $0.14 per million tokens. For context, OpenAI’s comparable offering was priced at $7.50 per million tokens at the time. That’s a 98% cost reduction.
The more recent V4-Pro model, which debuted in April 2026, comes in at $0.87 per million tokens for output and $0.435 for input. Those numbers represent a 34x reduction compared to leading rivals. The company has since locked in those prices permanently as of late May 2026.
Then there’s the V4-Flash variant, targeting budget-conscious workloads at around $0.14 per million tokens. That’s claimed to be 35x cheaper than GPT-5.5 alternatives. Third-party analyses have confirmed that DeepSeek models broadly run 20 to 50x cheaper than leading closed models.
The V4-Pro Max sits at the premium end of DeepSeek’s lineup with a blended price of roughly $2.17 per million tokens, confirmed by multiple benchmarking providers.
The V4-Pro’s performance benchmarks rival Claude Opus 4.6 and GPT-5.5, two of the most capable models on the market.
How they pulled it off
DeepSeek uses a Mixture-of-Experts (MoE) approach. Instead of activating all model parameters for every query, MoE architectures route each request to only the most relevant subset of the model’s expertise. This means less compute per query, which means lower costs per query.
DeepSeek’s previous V3 model reportedly cost $5.576 million in GPU rentals to train, compared to the hundreds of millions (and sometimes billions) that US-based labs have been spending on training runs.
The model is also fully open-source under the MIT license, which means anyone can deploy it, modify it, or build on top of it without paying licensing fees.
Why crypto and broader markets should care
The most immediate market impact was felt back in January 2025, when Nvidia’s stock experienced notable declines as investors recalibrated their expectations about the future demand for high-end compute hardware.
For the crypto sector specifically, decentralized AI projects have historically been constrained by the economics of inference costs. When running a model costs $7.50 per million tokens, decentralized compute networks struggle to compete with centralized cloud providers. At $0.14 per million tokens, the math changes dramatically.
DePIN networks focused on GPU compute, projects like Render, Akash, and io.net, might see their value propositions shift. Cheaper AI reduces the premium users will pay for compute, while it expands the total addressable market by making AI-powered applications viable for a much wider range of use cases.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·