The disappearing AI middle class
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
OpenAI and DeepSeek made opposite pricing moves within 24 hours, fundamentally reshaping the AI model market. GPT-5.5 launched at double the price of its predecessor ($5/$30 per million tokens), while DeepSeek released V4-Pro at roughly one-ninth the output cost and V4-Flash even cheaper, both under MIT license with open weights. This has eliminated the comfortable middle tier that most coding agents relied on, creating two distinct economic clusters. Developers now face a binary routing decision: pay for OpenAI's integrated outcome-focused product or run open-weight infrastructure. The analysis covers architectural reasons behind DeepSeek's pricing (MoE design, MIT licensing, Huawei Ascend chip support), implications for agent harnesses becoming model-agnostic, the new viability of self-hosting, and the weakening of Nvidia's monopoly on frontier AI inference.
Table of contents
The 24-hour splitThe widening gap for AI costsWhat OpenAI is actually sellingWhat DeepSeek is actually shippingThe middle is thinning, not goneWhat this means for the harness layerWhat’s nextSort: