Click to zoom
What a Year for Open AI Models!
Wow, 2025 has been nothing short of groundbreaking in the artificial intelligence arena. As we bid farewell to this eventful year, I'm eager to share our refreshed tier list for open model builders and spotlight the top models that have made waves. Let's dive into why these open AI technologies are laying the groundwork for an exciting future.
Top Models of the Year
Let’s break down the standout players dominating the open model scene:
- DeepSeek R1 (@deepseek_ai): This model has truly shaken up the AI landscape with unmatched capabilities. Honestly, it’s becoming a bit of a household name in our circles.
- Qwen 3 Family (@AlibabaGroup): Emerging as the new gold standard for open models, Qwen has really raised the bar this year. It’s impressive how these advancements in open AI technology keep coming.
- Kimi K2 Family (@Kimi_Moonshot): Kimi has disrupted expectations, proving that outstanding models can indeed emerge from China, effectively challenging the idea that DeepSeek holds the crown exclusively.
Runner-Up Models and Honorable Mentions
Don’t overlook these models, which have made meaningful strides in 2023:
- MiniMax M2 (@MiniMax__AI)
- GLM 4.5 (@Zai_org)
- GPT-OSS (OpenAI)
- Gemma 3 (@GoogleAI)
- Olmo 3 (@allen_ai)
- Nvidia's Parakeet (@nvidia): A cutting-edge speech-to-text model.
- Nemotron 2 LLM, Moondream 3 VLM (@moondreamai), Granite 4 LLMs (@IBMResearch), and HuggingFace's SmolLM3 (@huggingface) deserve a shout-out for their contributions too.
Updated Tier List for Open Model Builders
Here’s our revamped tier list reflecting the progress made this year:
- Frontier Open Labs: DeepSeek, Qwen, and Kimi Moonshot are leading the pack.
- Close Behind: The likes of Z.ai and MiniMax AI — both based in China — are showing us a shift in the global narrative away from U.S. dominance.
- Noteworthy: This tier celebrates innovation from both the U.S. and China, showcasing models from famous names like StepFun AI, Ant Group’s Inclusion AI, Meituan, Tencent, IBM, Nvidia, Google, and Mistral.
Predictions for 2026
Looking forward, here’s what I expect we’ll see in the open AI models arena by 2026:
- We’ll likely see ongoing scaling of open models as the demand continues to rise.
- The ongoing discussion about open model safety probably won’t undergo drastic changes.
- Expect more sectors to get engaged in developing open models.
- Trends such as Mixture of Experts (MoEs), hybrid attention mechanisms, and advanced fine-tuning methodologies will come into sharper focus.
- The gap between open and closed models should stabilize when viewed against public benchmarks.
- And here’s a hot take: no new Llama-branded open model releases from Meta are expected in 2026.
In a nutshell, the evolution of open AI models this year has not only redefined standards but also hinted at significant growth in the years ahead. The landscape keeps shifting, and staying informed about AI model advancements is essential for everyone involved.
If you’re hungry for more in-depth insights on the future of AI, resources on Forbes are a great start, or check out a video overview of these exciting developments on YouTube.
Thanks for reading!
If you found this article helpful, share it with others