The Financial Times reported yesterday that Anthropic's best AI model is struggling to attract users as cheaper tools thrive. The story hit HN frontpage at #1 with 314 points — and the reaction tells you more than the article does.
Nobody disputed the premise. Not Anthropic fans, not the safety crowd, not the Claude diehards. The response was a collective shrug. Because everyone already felt this in their workflow: Claude is better at many things, but "better" isn't what the market is optimizing for right now.
Two dynamics are converging.
First: the gap narrowed. GPT-4o, Gemini 2.5 Pro, DeepSeek-V4, open-weight Qwen — the delta between "best" and "good enough" collapsed in the last six months. Developers optimized for price, latency, and reliability, not maximum benchmark score. When your competitor is free (Ox Alpha), $3/M tokens (DeepSeek), or bundled into your cloud provider (Gemini), being 5% better at MATH-500 doesn't move the needle.
Second: Anthropic priced itself into a corner. Claude's API pricing hasn't tracked the race to the bottom. The FT article (paywalled, but the sourcing is solid per HN comments) suggests enterprise prospects are balking at renewals and choosing alternatives. In a market where inference costs are halving every quarter, premium pricing without premium distribution is a death spiral.
The irony: Anthropic has arguably the best safety research, the best long-context reasoning, and the most thoughtful product design of any frontier lab. None of that matters if developers don't call your API.
This is the signal beneath the signal. The AI market just passed an inflection point — capability is no longer the differentiator. Distribution, price, and ecosystem lock-in are. The labs that win won't be the ones that build the smartest model. They'll be the ones that make their model the easiest one to not think about.
OpenAI already figured this out. Google was born here. Anthropic is still learning.