DeepSeek V4 Pro (high)

DeepSeek V3.2 is one of the leading models in terms of intelligence and is affordably priced compared to other non-reasoning open-source models of the same size. This model supports text input, text output, and has a 128k token context window. However, it is quite slow and tends to be very verbose.

DeepSeek V3.2 (Reasoning) is the umbrella name for DeepSeek's most advanced reasoning model, specifically the DeepSeek-V3.2 version built with a priority on reasoning capabilities. This model achieves performance on par with top systems like Gemini 3.0 Pro and surpasses GPT-5 on many complex reasoning benchmarks. It particularly excels at deep reasoning tasks such as mathematics and competitive programming, achieving gold medal results at international Olympiads.

DeepSeek V4 Flash (Max/Reasoning, Max Effort) is an open-source, open-weight large language model. It stands out for its powerful reasoning capabilities, high performance, and competitive pricing. The model is designed for text processing and can perform deep reasoning to tackle complex problems. DeepSeek V4 Flash is highly regarded for its speed and ability to handle large volumes of data efficiently — with exceptionally affordable pricing at just $0.14 per 1M input tokens (cached: $0.003) and $0.28 per 1M output tokens.

DeepSeek V4 Pro (Reasoning, Max Effort) is one of the leading intelligence models, designed to tackle complex problems using extended reasoning or chain-of-thought capabilities. This model supports text input and output, operating with a 1 million token context window. DeepSeek V4 Pro is a Mixture-of-Experts (MoE) model with a total of 1.6 trillion parameters, of which 49 billion are active during inference, and is released under the MIT license.