DeepSeek
DeepSeek V3.1
DeepSeek-V3.1 is a hybrid model that supports both thinking mode and non-thinking mode. DeepSeek-V3.1 is fine-tuned on top of DeepSeek-V3.1-Base, which is built upon the original V3 base checkpoint using a two-phase long-context extension approach, following the methodology outlined in the original DeepSeek-V3 report. We have expanded our dataset by collecting additional long documents and substantially extending both training phases. The 32K extension phase has been increased tenfold to 630 billion tokens, while the 128K extension phase has been extended by 3.3 times to 209 billion tokens.
DeepSeek-OCR 2
DeepSeek-OCR 2 is a multimodal document recognition model released by DeepSeek AI, serving as an upgrade to DeepSeek-OCR. By introducing the DeepEncoder V2 architecture, it achieves a paradigm shift in visual encoding from "fixed scanning" to "semantic reasoning." The model replaces the original CLIP encoder with a lightweight language model (Qwen2-0.5B) and incorporates a causal flow query mechanism, while retaining the DeepSeek-3B-MoE decoder. The model requires only 256 to 1,120 visual tokens to cover complex document pages. On the OmniDocBench v1.5 benchmark, it achieves an overall score of 91.09%, a 3.73% improvement over its predecessor, with the reading order recognition edit distance reduced from 0.085 to 0.057.
DeepSeek R1 0528
DeepSeek R1 0528 is the latest open-source model released by the DeepSeek team, boasting impressive reasoning capabilities; notably, it achieves performance comparable to OpenAI's o1 model in mathematics, coding, and reasoning tasks.
DeepSeek V3 0324
DeepSeek V3, a 685B-parameter mixture-of-experts model, is the latest iteration of the flagship chat model family from the DeepSeek team.