DeepSeek V4 Pro undercuts the Western frontier by an order of magnitude

DeepSeek shipped V4 Pro on 2026-04-24: 1.6 trillion parameters, 49B active per token, 512K context. At $2.10 input / $4.40 output / $0.20 cached per million tokens, it sits in the same intelligence band as Claude Opus 4.7 ($5/$25) and GPT-5.5 Pro ($30/$180) — but at one-tenth to one-fourteenth the bill.

On TokenFloor the disparity is stark. A 100k-input / 5k-output agentic call runs $0.23 on DeepSeek V4 Pro vs $0.63 on Claude Opus 4.7 (before the +35% tokenizer tax) and $3.90 on GPT-5.5 Pro. Together AI hosts the same V4 Pro weights for US-billed buyers at parity pricing, so geography isn't a blocker.

The pricing pressure is structural: four Chinese open-weights frontier models landed in twelve days in May 2026 — DeepSeek V4, MiniMax M2.7, GLM 5.1, Kimi K2.6 — all under one-third of Claude Opus pricing. Vendor-side margin on the West-coast flagships looks increasingly hard to defend. Use the matrix sort to find the floor.