OpenAI Caught Between Claude and DeepSeek

If I want my fence painted, I don’t hire the world’s most intelligent person for the job. Firstly, because they’ll be expensive. But mainly because the odd-job guy down the street will probably do a better job anyway.

That’s the flaw in OpenAI’s pursuit, and hype, of super-human intelligence. Not just that they’re unlikely to achieve it with their current approach, but because most of the tasks we actually use AI for don’t need super-human intelligence in the first place.

DeepSeek and the rest of the Chinese AI providers seem well aware of this issue. Last week, DeepSeek released a model that may not match ChatGPT 5.4’s capabilities, but it is more than good enough for 99.9% of what ChatGPT 5.4 is actually used for while being massively cheaper.

On the other end of the market, Anthropic’s Claude is quietly conquering the high-end developer and software economy. Anthropic’s models aren’t getting much more “intelligent” in the abstract, AGI-chasing sense. But they are getting better at the narrow tasks like writing code, using tools, planning, better output formats, and agentic workflows.

OpenAI is caught in the middle.

Front #1: Anthropic Owns the Software Economy

OpenRouter provides us with some insight into the most popular models and it is currently showing the first OpenAI model at 14th place – and these figures do not include the new DeepSeek V4 model.

Trackers like Similarweb show OpenAI’s traffic share falling, but even casual conversation reveals people that only knew “ChatGPT” six months ago now have Claude Cowork, Claude Code or Claude Designer accounts.

Front #2: DeepSeek et al come for the Volume Market

While Anthropic eats the premium tier, China’s DeepSeek has just released V4, a model that threatens to fundamentally break the economics of the American AI industry.

DeepSeek V4 delivers performance close to leading US models while costing **55% to 95% less**. According to Artificial Analysis, the **DeepSeek V4 Flash** sits firmly in the “most attractive quadrant” of price vs. performance:

• Roughly **50% cheaper** than Google’s Gemini 3.1 Flash Lite
• Up to **95% cheaper** than Anthropic’s Claude Sonnet 4.7
• A **1-million token context window**
• Agentic capabilities rivaling GPT-5.5 and Opus 4.7

And here’s the kicker: it was trained on “nerfed” hardware due to US export controls. The Chinese labs are innovating around the algorithm, not the silicon.

This won’t be another “DeepSeek moment” that shocks markets overnight. It’s a slow, steady commoditization of the middle of the market, where most enterprise workloads actually live.

The Bottom Line

ChatGPT still has the brand, but it’s slowly being tied to the most simple forms of chat, whereas Claude is tied to getting stuff done, something that people and companies will pay for.

Anthropic is winning where the money actually flows, through APIs, into software products, embedded in the tools developers use every day. DeepSeek is winning where the volume actually lives, in cost-sensitive enterprise workloads that just need the job done.

OpenAI needs to stop chasing its own hype and start getting practical. Because right now, both ends are being pulled out from under it.