Which AI Model Do I Use for What? The Full List (2026)
The short answer
I don't use a single AI; each job has its own model. Code and video editing run on Claude Code (Opus 5.5), my talking assistants use OpenAI's live voice model GPT-Live, phone calls go through a NetGSM line and an Asterisk PBX with GPT-Live speaking, the security camera sees with YOLO11 and YuNet, images come from GPT Image 2.5, logos from Nano Banana 2.1, video from Veo 3, music from Suno, speech-to-text from Whisper, research from Gemini Flash, Instagram data from Apify, and social media planning plus comment automation from Postuby. Jarvis thinks with Claude Sonnet 5.5 and my website chatbot runs on Claude Haiku 5.5. I pay for all of it myself; Gurizon and Postuby are my own companies.
Code and talking assistants
Everything you see in my videos follows the same pattern: I describe what I want, Claude Code writes the code, and OpenAI's live voice model does the talking. The table shows which model runs where.
| Job | Model | Where you saw it |
|---|---|---|
| Writing code, building apps and sites | Claude Code (Opus 5.5) | Jarvis, Irgat, the balcony radar, this site, most of Postuby |
| The voice (Gurizon Assistant, Jarvis, Irgat) | OpenAI GPT-Live, decisions by GPT-5.4 mini | The assistant that said "you can't talk to me like that" and shut the screen |
| Phone calls (barber booking, haggling, calling my editor) | NetGSM 0850 line + Asterisk PBX + GPT-Live | The assistant-calls-the-barber videos |
| Jarvis (desktop assistant) | Claude Sonnet 5.5 thinks, GPT-Live talks, small jobs on Gemini Flash | The "it calls me boss" video; I give it away for free |
| Irgat, my 8-year-old security camera | YOLO11 objects, YuNet faces, OpenCV; voice by OpenAI | The camera complaining about overtime, the camera watching our PlayStation match |
| Balcony radar | YOLO11 + OpenCV, my own Python | The "my wife caught me" video |
| The chatbot on this site (my digital twin) | Claude Haiku 5.5, with Gemini 3.8 Flash and GPT-4o as backups | The chat in the bottom right |
| Laf Sok (comeback writer) | Gemini 3.5 Flash Lite, smart mode on Gemini 3.8 Flash | The AI that talks back in the WhatsApp group |
| The hologram look | My own code, in the browser | The assistant's sphere on screen |
I don't write code. I built Jarvis, Irgat, the radar and this site by describing them to Claude Code. The hard part isn't the code, it's saying clearly what you want and checking the result.
Images, video and sound
| Job | Model | Note |
|---|---|---|
| Image generation (Reels covers, Postuby visuals) | GPT Image 2.5 | My real photo is the base; it builds the scene |
| Logo generation | Nano Banana 2.1 | Cleaner lettering and lines for logos |
| Video generation | Veo 3, restyling with Gemini Omni | I generate 8-second scenes and stitch them |
| Music, sound effects, the Rahatla and Tesla sounds | Suno + my own 3D audio scripts | The brand voice at the end of videos is ElevenLabs |
| Speech to text | Whisper large-v3-turbo | Runs on my Mac; subtitles and cut points come from it |
| Editing the videos | Claude Code + ffmpeg + Chrome | Cuts, audio cleanup, big-head animations, subtitles, effects |
| Uploading to YouTube | A Mac app I wrote | I call the assistant and say "upload it" |
The videos I call "edited by AI" come out of exactly this chain: the footage, a Whisper transcript, editing code written by Claude Code, and ffmpeg.
Data and social media
| Job | Model / tool | Note |
|---|---|---|
| Research, current news | Gemini Flash (with Google Search) | Source checks for news videos; second opinion from Claude's web search |
| Pulling Instagram data | Apify | Hourly views and follower trends; Claude does the analysis |
| Social media planning, publishing, comment-to-DM | Postuby (my own company) | Runs GPT-5.4 nano and mini, GPT-5 and Gemini 3.5 Flash inside |
| One gateway to image, video and music models | Kie | I call GPT Image, Nano Banana, Veo and Suno from one account |
How the pieces are wired
Camera
go2rtc pulls the camera stream, YOLO11 finds objects and YuNet finds faces, GPT-5.4 mini decides what to do, GPT-Live speaks, and it comes out of the speaker.
Phone
The assistant hands the call to the Asterisk PBX, the PBX dials out over the NetGSM line, and GPT-Live does the talking. The other side is a real person; nothing is staged.
Someone comments a keyword, Postuby sees it, checks whether they follow me, and sends the link by DM.
Video
The footage goes into the Mac, Whisper transcribes it, Claude Code writes the editing code, ffmpeg renders the video, and the Mac app uploads it to YouTube.
Why not a single model?
Because each model is good at a different job. Claude makes the fewest mistakes on long code and editing. OpenAI's live voice model stays natural even when you interrupt it. Gemini with search is fast and cheap for checking news. GPT Image is more realistic for photos, Nano Banana is cleaner for logos. I try two models side by side on a job and keep the one that wins, which is why this list changes.
Disclosure: Gurizon and Postuby are my own companies. I have no partnership with any other tool on this list; I pay for all of them myself.
If you want to try them
- Laf Sok is free: safaktozar.com.tr/laf
- Rahatla, the 3D sound session: safaktozar.com.tr/rahatla
- Jarvis, the desktop assistant: safaktozar.com.tr/asistan
- Step-by-step guides for the camera and phone setups: the How-To section
Key takeaways
- There is no single AI; there is a model per job. Code and editing on Claude, speech on OpenAI GPT-Live, images on GPT Image, logos on Nano Banana, video on Veo 3, music on Suno.
- The phone calls are real: a NetGSM line, an Asterisk PBX, GPT-Live speaking.
- The camera and radar don't send video to the cloud; YOLO11, YuNet and OpenCV run locally.
- The whole social media side lives in Postuby: planning, publishing, comment-to-DM automation.
- The list changes as new models ship; check the date.
Frequently asked
Which one do you pay the most for?
Claude. All the code, editing and analysis happen there; OpenAI's voice model comes second because the assistants are billed per minute of talking.
Is Jarvis really free?
The app is free. Voice chat runs on your own OpenAI key, and that usage is paid to OpenAI.
Does the camera send conversations to the cloud?
The video stays on the computer. The talking part runs on OpenAI's voice model, so audio does go to the cloud. The Whisper transcripts for my videos are fully local.
Where should a beginner start?
With one chat model: ChatGPT, Claude or Gemini. Work with it alone for a month; once you see where it falls short, add a second tool for that job. Don't try to build this list in a day.