The Best Open Weights Coding Models of 2025 (blog.brokk.ai)

DeepSeek has launched what is anticipated to be the final major open coding model of 2025, prompting a reflection on the current state of AI coding models. The latest performance rankings reveal that while local models are inching closer to being viable for specific use cases, they still fall significantly short of the capabilities offered by larger models. The top open-weights models, such as GLM 4.6 and DeepSeek-V3.2, are now only about six months behind their closed counterparts in terms of intelligence, offering a cost-effective alternative to more expensive closed models like GPT-5-Mini. However, a critical challenge remains: speed. Open models are currently slower than their closed-lab contemporaries, primarily due to differences in size, optimization of inference infrastructure, and capacity constraints. As the AI/ML community looks ahead, maintaining the pace of development in intelligence is essential, but improving efficiency is becoming the next significant hurdle. If open labs can enhance their models to match the speed of cloud solutions while retaining their competitive intelligence by 2026, it could signal a transformative era where open models finally hold their own in terms of performance against the industry's giants.