What do you think about current AI models?

[tzxqjai]

🏷️ Discussion Type

Question

💬 Feature/Topic Area

API

Body

Lately I’ve been using a lot of LLMs for development, scripting, and AI comic work.
From OpenAI and Claude to open-source models like DeepSeek, Qwen, and others, the progress is really fast.
But I also feel some pain points clearly:
Official APIs are getting expensive for daily batch use
Region locks and access issues still exist
Long-context performance varies a lot between models
Stability can be inconsistent for production
I’m curious about everyone’s experience:
Which model do you prefer for real projects?
What do you think is the biggest strength and weakness of today’s AI models?
What improvements do you expect to see next?
Looking forward to your thoughts.

[Mehrab-Javadzadeh]

I do not think there is a best model anymore. It really depends on what you're trying to do with the model. For coding I value reasoning and consistency in the model more than the benchmark scores of the model. Good context handling is also important for the model especially when you are working on projects with the model.
The biggest strengths of the model are how much the model helps with coding, debugging and brainstorming with the model. The biggest weakness of the model is that the model can still sound confident even when the model is wrong.
I would love to see consistent reasoning in the model better support for large codebases in the model lower API costs for the model and easier switching, between different providers of the model. It will be interesting to see how much things improve over the year with the model.

[Beno-Goulart]

I think tools like OpenCode and Ollama show an important direction for the future of AI development.

Ollama makes it much easier to run open-source models locally, such as Llama, Qwen, and DeepSeek, giving developers more control over privacy, costs, and experimentation without depending entirely on expensive APIs or region availability.

Combined with OpenCode, this creates a more flexible coding workflow where developers can use AI assistance while keeping more of their environment and data under their control. The main limitation is that local models still depend heavily on hardware and may not always reach the reasoning quality of the most advanced cloud models.

I believe the future will likely be a hybrid approach: using powerful cloud models for complex tasks and local solutions like Ollama + OpenCode for everyday development, privacy-sensitive projects, and reducing operational costs.

[Sarveshero3]

I use groq apis mostly for like the llm integration in my project otherwise like for AI-assisted coding or vibe coding, i just use antigravity, Opencode. Yes, its dogfight out there to get a better model and more tokens but thats all i do. Rotate your accounts like the burner ones and use it on daily basis, make sure that you are not wasting any time and use the moment you got them otherwise you miss the opportunity to use them the moment you need it the most.
Strength are that too powerful and understands things with far less effort as before but only with good prompts and contexts otherwise the weakness is that they all are too expensive and we are becoming too dependent on them whether we like it or not.
Next development is still in an ambiguous position, i honestly don't know the next big thing for llms, they say its AGI but that's the end goal but for me it might something like better agent conversation, all the agents we converse are too straightforward and day by day it is changing and cross questioning us but i feel that part can be better. Take all of this with a grain of salt, Thank you

[imdaniissh]

Current AI models are incredibly useful for coding, research, and productivity. While they have become much more capable, I believe improving accuracy, reasoning, affordability, and long-term context handling will make them even more reliable for real-world applications.