DeepSeek launches V4-Pro and V4-Flash with dual API support

DeepSeek announced support for two new models in its API: DeepSeek-V4-Pro and DeepSeek-V4-Flash, available as of April 24, 2026. These models can be invoked via both the OpenAI ChatCompletions interface and the Anthropic interface, with the existing base_url unchanged. Developers simply set the model parameter to deepseek-v4-pro or deepseek-v4-flash. The V4 family represents a significant step forward, offering enhanced reasoning and agent capabilities over previous versions. The dual-interface support ensures a seamless migration for teams already using either ecosystem, reducing integration overhead.

To smooth the transition, the legacy model names deepseek-chat and deepseek-reasoner will be deprecated on July 24, 2026 — a three-month grace period. During this time, deepseek-chat maps to the non-thinking mode of deepseek-v4-flash, and deepseek-reasoner maps to its thinking mode. This allows existing applications to continue working without immediate code changes while giving teams time to update their model parameters. The changelog also traces the evolution from V3.2 through V3.1 and R1, showing a consistent focus on reasoning benchmarks, agent tool use, and hallucination reduction. DeepSeek recommends that all users migrate to the explicitly named V4 models to benefit from the latest optimizations and avoid service disruption after the cutoff date.