RISK PROFILE COMPARISON — JULY 28, 2026
● Active US sanctions threat: Moonshot AI only — DeepSeek has no current government accusation
● Distillation accusation: Moonshot only — accused of distilling Fable 5 to build K3
● API price: DeepSeek V4 Pro $0.44/M (cheapest capable API) vs K3 $3/M
● Intelligence level: K3 leads — AA Index #4 (score 57) vs V4 Pro not ranked at frontier
● Both: China NI Law on hosted API (resolved via Western self-hosting), FLI effectively failing grade
● Self-hostable: Both — V4 Pro weights available, K3 weights live July 27
Both DeepSeek and Moonshot carry China NI Law data residency risk on their hosted APIs, resolved by self-hosting on Western infrastructure. But the sanctions risk is not symmetric: the US government has made specific accusations against Moonshot that it has not made against DeepSeek. For teams evaluating which Chinese open-weight model to use, that asymmetry matters — DeepSeek V4 Pro at $0.44/M is the lower-regulatory-risk Chinese option today, at the cost of significantly lower intelligence ceiling than K3.
DeepSeek V4 Pro for: Cost-sensitive workloads where intelligence ceiling is not the constraint, and where the lower regulatory profile (no active sanctions threat) matters. Best Chinese open-weight option on regulatory risk today.
Kimi K3 (self-hosted) for: Teams that need frontier-tier intelligence from an open model, have already downloaded weights under Modified MIT, and have received legal sign-off on self-hosted deployment pending sanctions resolution. Freeze new Moonshot API commitments until outcome is clear.
Last updated July 28, 2026. Related: Moonshot sanctions threat → · DeepSeek V4 Flash vs V4 Pro →