DeepSeek V4-Pro: 1.6T params, 1M-token context, sharply lower inference cost — reportedly forces Anthropic price cut

DeepSeek released V4 via API on April 24, 2026, with a Pro variant carrying 1.6 trillion total parameters and a 1 million token context window. V4-Pro requires substantially less inference FLOPs and KV-cache than V3.2, making it competitive for agentic loops and large-codebase workloads. The release has stirred AI economics, with Anthropic reportedly cutting Opus 4.7 prices in response, and DeepSeek separately optimizing the model for Huawei chips.