Deepseek's Code2.0 model set to challenge AI leaders with 3 trillion parameters

DeepSeek, a prominent player in the AI field, is set to launch its Code2.0 model, which is rumored to have over 3 trillion parameters. This model is expected to be unveiled in September 2026, marking a significant leap from its previous offerings. The company has already released V4.1Flash, which features a new architecture and is considered a generation update. The V4.1Flash reportedly achieved a doubling of parameters compared to its predecessor, setting the stage for Code2.0's 3 trillion parameter count. Industry sources suggest that Code2.0's focus on 'computer control' will allow AI to automate complex tasks for users. The model is anticipated to outperform existing frontier models like Mythos5.1 and GPT-6Astra. DeepSeek's strategy of releasing multiple product lines, including the V4.1Pro series and Code2.0, continues to push the boundaries of global AI capabilities. The company's recent release of V4.1Flash, which represents an architectural overhaul, has raised market expectations for its subsequent products. While the V4Pro-0813 version did not meet initial performance benchmarks, the market remains cautiously optimistic about the V4.1Pro series. DeepSeek's Code2.0 is positioned to compete with models like K3's 2.8 trillion parameter model and Qwen3.8Max's 2.4 trillion parameter model. The company's multi-route product strategy, including the release of Code2.0, is seen as a key factor in its ongoing efforts to lead in the AI industry.