Updated August 2026: V4 Flash moved to the 0731 public beta build on July 31, 2026. This rewrite includes the latest model IDs and pricing.
DeepSeek’s V4 generation is the price-performance story of 2026 — and the company sits at the center of the year’s biggest AI controversy. Here’s the full picture.
The Current Lineup
Two models anchor the family: V4 Flash at $0.14/M input and $0.28/M output (284B total / ~13B activated MoE, 1M context, MIT open weights), and V4 Pro at $0.435/$0.87 for stronger reasoning. The July 31, 2026 update (V4 Flash 0731) reruns post-training on the same architecture — same price, refreshed behavior. A future 2x peak pricing schedule has been announced.
Migration Deadline Passed
The legacy deepseek-chat and deepseek-reasoner endpoints were fully retired July 24, 2026. All new integrations should use deepseek-v4-flash and deepseek-v4-pro directly.
The Distillation Shadow
Anthropic suspended DeepSeek (with MiniMax and Moonshot) for alleged data distillation — 24K fraudulent accounts, 16M interactions. DeepSeek denies wrongdoing. The controversy doesn’t change the model’s capability, but it may affect how the US government treats the company — and whether US enterprises adopt it.
What It Means for Buyers
DeepSeek V4 Flash remains the strongest price-per-token open-weights play, now with a current build ID and MIT licensing. Watch for the 2x peak pricing schedule and the resolution of the distillation dispute.
Context: AI Models in 2026: Complete Guide
