Can anyone suggest which is the best
3 Likes
Personally I prefer to steer clear of closed source AI, so the open source models are more appealing to me.
Try the most popular ones and see which you prefer? 
3 Likes
I didn’t know about Qwen, Kimi and Mistral. 
2 Likes
It’s too early to say but in the last week I’ve:
built a CoT system with the help of ChatGPT and a Deep Research system with DeepSeek V3.
I deleted my Qwen account because PAYG didn’t mean to them what it meant to me.
Revived Cohere’s websearch connection for use with the Deep Research (DS) project. We’re going generic 
Shelved the Gemini 2.0 Thinking function because it no longer delivers the CoT.
Topped up my Mistral balance because on balance I’m moving away from US AI models.
3 Likes
Here is something else to play with as a developer, the Zed editor. It lets you pull in files, folders, or active tabs for context. It supports a wide variety of different AI interfaces.
I’ve also started looking at the Gemini 2.0 Flash model. In the context of development work, the huge context size is interesting. Theo had a YT video about it.
2 Likes
I’ve enjoyed Cline, which integrates best with Anthropic’s Claude 3.5 Sonnet model. It’s very smart, as long as your language, fromework, and desired results are all reasonably common / popular. Sadly, Anthropic is the priciest of the models.
1 Like
DeepSeek and ChatGPT both have strengths for developers. DeepSeek is optimized for code generation and completions, especially for specific programming tasks. ChatGPT offers a broader range of AI-assisted coding help, debugging, and explanations with a more conversational approach. If you need precise code generation, DeepSeek might be better, but for general problem-solving and learning, ChatGPT is more versatile.
1 Like
For developers, there are benefits to using different types of AI. A fixed price subscription will allow for unlimited reasoning about what you want to accomplish. Using different AI providers and/or models can broaden the scope of your results as each will reason in a different way. Then, a very well integrated pay per token AI can have much greater context for actually implementing changes. I wrote a blog post about my experience doing this with a combination of ChatGPT and Claude Code.