ChatGPT was the first AI companion that helped get this particular creative universe moving. That earns it the first subdomain. It does not earn it first place in every benchmark.
This is not a tribute to a winner. It is an appreciation of an extraordinary technological moment, the people and organisations building it, and the human judgment needed to use it well.
No affiliation with OpenAINo chatbot · No API · No tracking scriptsSources linked throughoutIndependent Artificial Analysis Intelligence Index v4.3.2 scores, as reported in an October 2026 snapshot. Higher is better on this composite; these are index points, not accuracy percentages or IQ scores. Model settings, prices, and rankings change.
Bars use a 0–60 display scale. These scores represent specific tested model configurations, not necessarily the default free consumer chat experience. Source: Artificial Analysis primary benchmark report; see also the live Artificial Analysis leaderboard. Snapshot date: October 2026. Rankings are not permanent.
On a different benchmark, SWE-bench Live reported 71.0% resolved for a DeepSeek-powered agent and 70.3% for a GPT-5.6-powered agent in a 300-issue evaluation. That measures agent-plus-model software repair, not the ability of either chat product to write an essay or have a conversation. Inspect the methodology and leaderboard →
Particularly useful as a general-purpose conversational workspace: drafting, explaining, coding, analysing files and data, generating images, and iterating through ideas. Its breadth is a strength; it is not immunity from confident mistakes.
Deserves serious attention for substantial writing, code, and complex task execution. In the cited composite benchmark snapshot, its leading models outscore the cited GPT and Gemini models. That is a measured advantage on this evaluation, not universal superiority.
A strong option when the task benefits from Google's ecosystem, multimodal capabilities, and its own model family. Some tasks will favour Gemini, but any claim that it universally wins should be tied to a named test and configuration.
Its distinctive product emphasis is sourced web search and research, rather than a claim to own the single best underlying model. Citations make investigation easier; they still need to be opened, read, and checked.
These are practical workflow distinctions, not a controlled four-product experiment. Products and subscriptions change; test with your actual task before committing.
The remarkable change is not that the machine has become an artist, researcher, developer and editor all at once. It is that one person can now coordinate assistance across all those disciplines.
Research can become a reading map; notes can become an essay; an essay can become a web page; a web page can become a public experiment. The same individual can move among roles that once demanded separate software stacks and specialist teams. That expands access to production. It does not guarantee originality, correctness, taste, scientific validity, security, or an audience.
The work still needs a human who chooses the question, notices when an answer is hollow, checks the citation, tests the code, edits the prose, and decides what is worth putting into the world. AI can compress production friction; it cannot automatically supply judgment.
Models can fabricate facts, references, and explanations while sounding persuasive. Even OpenAI explicitly documents this limitation. Read OpenAI's own warning ↗
A coding-agent result, an intelligence index, and a web-search product measure different things. Compare like with like; evaluate your own work rather than outsourcing every decision to a leaderboard.
Publishing faster is useful only when the work remains thoughtful. A studio-level workflow is newly accessible; studio-level output must still be earned through revision and testing.
Models can help people articulate ideas, but lived experience, values, relationships, accountability, and responsibility remain human.
This page exists inside a growing, independent set of experiments in writing, education, consciousness, organisational behaviour, and perfumery. The sites are examples of work enabled by a human-plus-tools workflow, not evidence that a particular model is uniquely responsible for their quality.
Essays, ideas, experiments and the wider personal website.
Education, reflections on consciousness, and a creative perfume project.
Related conceptual projects, each with its own frame.
Use ChatGPT when it helps. Switch to Claude when Claude does better. Bring in Gemini, Perplexity, or another tool when the problem calls for it. Loyalty belongs to the quality of the work, not the logo.