Skip to content
MJ Marketing

AI in Numbers

AI isn't a buzzword here. Here's the proof.

For me AI is a tool, not an end in itself — used deliberately so your campaigns are built faster, analysed more thoroughly and stay technologically future-proof.

1.1B
tokens processed
2,582
images generated
20
AI models
33
AI tools

As of: July 01, 2026 · since 2023

Why this matters for you

01

Speed

Research, setups and implementations that used to take days now run in hours — that time goes into your results, not busywork.

02

Depth

More analysis per account: keyword research, ad copy and reporting at a thoroughness that wouldn't be economical by hand.

03

From practice, not a slide

When I advise you on AI, I only recommend what I run in production every day — no hype deck.

Models

Tokens by model

One model dominates, many complement it — here's the logged token volume per model.

722M
113M
95.7M
73.9M
31.8M
22.1M
13M
9.9M
8.2M
4.3M
1.6M
746K
GPT-5.5
Qwen3 Coder 480B
Horizon Beta
Claude Opus 4.8
Kimi K2
Horizon Alpha
Qwen3 235B A22B
Claude Sonnet 4
Claude Sonnet 4.6
Sonoma Dusk Alpha
Nemotron Nano 9B V2
GLM 4.5 Air

Bar height on a logarithmic scale — the numbers shown are exact.

+ 8 more models in the long tail

The stack

What I work with every day

Coding agents & IDEs

Tools that write code, refactor and ship whole features — partly autonomous as agents, partly right inside the IDE. They build campaign setups, tracking implementations and this very website.

Claude Code
Claude Code
Daily driver for agentic coding — the main engine.
Codex
Codex
OpenAI's coding agent, the counterpart to Claude Code.
Gemini CLI
Gemini CLI
Gemini in the CLI and as a VS Code extension.
Kilo Code
Kilo Code
Agentic coding across many models — used very often.
Cursor
Cursor
AI-native IDE.
Kiro
Kiro
Agentic IDE.
Roo Code
Roo Code
Coding agent, similar to Kilo Code.
Cline
Cline
Coding agent inside the IDE.
Antigravity
Antigravity
Google's agentic IDE — used intensively.
AI Studio
AI Studio
Google AI Studio — building dynamic web-app systems.

Models & providers

The actual AI models behind the tools — from OpenAI, Anthropic and Google to open models like Qwen, DeepSeek or Gemma. Each provider has different strengths; I pick the right model per task instead of betting on a single one.

OpenAI
OpenAI
GPT models & Codex.
Claude
Claude
Claude (Opus/Sonnet) for analysis & coding.
Gemini
Gemini
Gemini (incl. Nano Banana) — used a lot.
Gemma
Gemma
Google's open models — fast, for AI agents.
Qwen
Qwen
Alibaba's Qwen — local & via API/OpenRouter, incl. coding.
Kimi
Kimi
Kimi K2 (Moonshot) — strong for agentic coding.
DeepSeek
DeepSeek
DeepSeek — coding & agentic tasks.
Grok
Grok
Grok (xAI) — tested, incl. cloaked as “Sonoma” on OpenRouter.
MiniMax
MiniMax
MiniMax — cheap, good for quick writing tasks.
NVIDIA Nemotron
NVIDIA Nemotron
Nemotron — tested for AI agents.
Z.ai (GLM)
Z.ai (GLM)
GLM models from Z.ai (Zhipu) — tested for coding & agents.

Image · video · audio

Image, video and audio generation — locally on my own hardware (ComfyUI) and via providers. For ad creatives, mockups, blog headers and voiceover, without expensive stock rounds.

Midjourney
Midjourney
Image generation for creatives & headers.
GPT Image
GPT Image
OpenAI's image generation in ChatGPT (GPT Image) — creatives & edits.
Nano Banana
Nano Banana
Google's Gemini image generation (Nano Banana / Pro) — fast creatives & edits.
ComfyUI
ComfyUI
Local image, video and audio (TTS) generation.
Sora
Sora
Video generation (not for professional use).

Design, automation & research

From wireframes through process automation to research agents — the tools that form the actual workflow around the models.

Figma
Figma
Design + Figma Make for quick wireframes & mockups.
Make
Make
Automations & workflows.
Manus
Manus
Agent platform — tested.
MCP
MCP
Model Context Protocol — connects tools & agents to the models.
Perplexity
Perplexity
Research — also for research agents.
Lovable
Lovable
App builder — tested.

Gateway

One access point, dozens of models: quickly testing and switching between providers without a separate account per model.

OpenRouter
OpenRouter
Model gateway — go-to for quickly testing new models.

How these numbers are produced

measuredfrom the providers' usage dashboards

The token figures come straight from each provider's usage dashboard (OpenAI, Anthropic, OpenRouter, Kilo), the image figure from the Midjourney account. It's the sum of logged usage up to the snapshot date — continuously growing. Stealth-tested models (Sonoma = xAI Grok, Horizon = OpenAI) are attributed to their confirmed maker.

AI that flows into your results

From the campaign to the tracking: the same toolbox works on your account. Let's talk about what that makes possible for you.

Direct contact with Mijo

Ready for real results?
Let's be direct.

No sales pitch, no pressure – just an honest assessment of whether and how Google Ads can work for your business.

Address

Olszanska 7, 31-513 Kraków, Polen

Availability

Mon-Fri: 9:00 AM - 6:00 PM

Book a discovery call

Pick a time that fits – the calendar shows my live availability.

Open full calendar
5.0 · 17 reviews on Google

You talk directly with Mijo – no account manager.