DeepSeek is making the 75% discount on V4 Pro API pricing permanent, locking in rates that undercut most competitors. The deepseek-v4-pro model will cost $0.003625 per 1M input tokens with a cache hit, and $0.87 per 1M output tokens after the promotional period ends on 2026/05/31. This permanent price cut follows a temporary discount that originally reduced costs by three-quarters.

The deepseek-v4-flash variant remains significantly cheaper, charging $0.0028 per 1M input tokens on cache hits and supporting up to 2,500 concurrent requests. V4 Pro maxes out at 500 concurrent requests. Both models handle 1M token context windows and 384K max output, with support for tool calls, JSON output, and beta features like chat prefix and FIM completion.

Deprecation notices warn that the older deepseek-chat and deepseek-reasoner model names will be phased out, redirecting to non-thinking and thinking modes in the newer V4 family. The aggressive pricing signals DeepSeek's intent to maintain low-cost LLM API access, especially as cache hit pricing now sits at one-tenth of launch rates across all models.