The open-source large language model (LLM) arena is witnessing a dynamic global competition, with significant contributions from companies worldwide, notably Meta and several Chinese tech giants. A recent social media post by Haider. suggested that the open-source competition is "mostly b/w Chinese companies," citing models like Qwen 3, Kimi K2, and DeepSeek R1, while claiming "Meta isn't in the race." However, current data indicates a more diverse and intensely competitive environment.
Chinese firms have indeed made substantial strides in the open-source LLM space. Alibaba Cloud's Qwen 3 models, released in 2025, are recognized for their strong performance, particularly in multilingual capabilities and code generation tasks, making them valuable for developers in AI applications. DeepSeek R1, from Chinese AI company DeepSeek, has also consistently ranked high on various leaderboards, with its latest iteration, DeepSeek-V3, topping the Chatbot Arena open-source leaderboard in May 2025, excelling in reasoning, mathematics, and code generation.
Contrary to the tweet's assertion, Kimi AI, developed by Chinese startup Moonshot AI, is notable for its exceptionally long context window. In March 2024, Moonshot AI upgraded Kimi AI's context window to an industry-leading 2 million characters, enabling it to process vast amounts of information for complex tasks, directly refuting the claim of being "limited by context size."
Meta Platforms, Inc. remains a formidable force in the open-source LLM domain. Its Llama 3 family, including the recently released Llama 3.1 in July 2024 with models up to 405 billion parameters, consistently ranks among the top performers on open-source leaderboards. These models are widely adopted and are considered foundational in powering advanced AI applications, indicating Meta's significant and ongoing presence in the open-source LLM development.
The global competition extends beyond these players, with other prominent open-source LLMs emerging from various regions. French startup Mistral AI, Google DeepMind's Gemma series, and Cohere's Command R/R+ models are also leading benchmarks and evaluation boards, showcasing a broad international effort in democratizing AI technology through open-source initiatives. The evolving landscape fosters rapid innovation and collaboration, driving advancements in the field at an unprecedented pace.