Unified API for AI Tokens

DeepSeek-V4-Pro is DeepSeek's flagship MoE hybrid expert model, employing a highly efficient architecture with 1.6T total parameters and 49B activation parameters, and natively supporting an ultra-long context window of 1 million tokens. The model features a dedicated inference mode, achieving industry-leading performance in three core areas: complex logical reasoning, professional code generation, and intelligent agent execution. Its overall performance rivals that of top-tier closed-source models globally.

Seedream 4.5 is Doubao's latest multimodal image model, integrating text-to-image, image-to-image, and image fusion capabilities, while incorporating common sense and reasoning abilities. Compared to its predecessor, the 4.0 model, it boasts significantly improved generation quality, better editing consistency and multi-image fusion effects, more precise control over image details, more natural generation of small text and faces, more harmonious image layout and colors, and enhanced aesthetics.

Kimi-K2.5 is the most versatile model released by Dark Side of the Moon to date, featuring a native multimodal architecture that supports both visual and text input, thinking and non-thinking modes, and dialogue and agent-based tasks.