What is current your "daily driver" coding model? (July 2026)

Fable 5 at Max effort? OpenAI's gpt-5.6? The latest Gemini models? Perhaps Kimi K3 or other open-weight models?

For your typical needs these days, what model do you tend to use and why?

Comments

[FrancisTRᴅᴇᴠ]

I use Gemma4 (Cloud Version) via Ollama. It's free to use and I can integrate it in my VScode without an issue. I tend to not go with models that are paid because they are...paid. I like saving money lol

[Paulo H. Alkmin]

On my home server, I like to have open-source models for daily usage (automations, reminders, etc.), and Gemma 4 works well.

Been testing Kimi K3 with Kimi CLI, and it's a great model, but token limits are brutal. Opus 5 is perfect for doing the hard work, and Gemini is ok-ish for minor tasks.

[Daniel Nwaneri]

Gemini does a lot of my day-to-day heavy lifting and gets maybe 10% of the credit it deserves. But the moment I need actual coding muscle, I'm back on Claude without hesitation. The gap there is still real and Gemini hasn't closed it.

[Randal L. Schwartz]

I'm using Antigravity, running flash 3.6 medium, and my multi-step workflow: gist.github.com/RandalSchwartz/598... which has been organically grown using lots of real-world iterations.

[Fayaz]

• For serious coding works: mostly no AI these days. Just my good old brain. Tried different models, but I don't get fulfillment that way. Also, I fear I'll lose my development ability. Not interested. I'm still working on to find a good balance between productivity gain using AI and keeping my dev ability when doing serious dev works. Still didn't find the sweet spot!
• For scripting (DevOps, automation etc.): Sonnet and GPT 5.6 mostly. Good enough.
• For 1st prototype / MVP: Sonnet, Opus, GPT-5.6, Gemini Flash.
• For learning, researching: NotebookLM, ChatGPT Deep Research.
• For Testing out AI capabilities, Hackathons: all sorts of new and old models, including open models.

[Vicente G. Reyes]

I think I'm using Opus 5. I used to use Antigravity's models but switched to Claude in March because I saw that it was the most efficient in all models I've used.

[Thorin Tabor]

Claude Sonnet for most targeted coding work, with a little bit of Opus mixed in for occasional architectural work. However, if I'm doing things in bulk I usually run locally using either Gemma 4 or Qwen.

[Kantemir Satibalov]

I usually use the cursor in automatic mode. I also use: qwen, Deepseek, kimi