Metamorfon integrates eight distinct providers — Anthropic, Google, OpenAI, Mistral, DeepSeek, Qwen, xAI, Cohere — where most tools on the market stop at one or two. This choice is not a matter of functional richness or commercial coverage. It follows from a methodological requirement without which Metamorfon would not do what it claims to do. And it carries, more broadly, a certain view of what the artificial intelligence ecosystem must demand to remain useful.
The value of a Metamorfon session rests on the epistemic diversity of the models confronted. This diversity does not reduce to the diversity of models from a single provider. Two models from the same provider, however different they may be technically, share training corpora, alignment philosophies, choices of guardrails, and response inclinations inherited from a single engineering culture. Their confrontation produces nuances, sometimes valuable ones, but rarely points of friction that are genuinely structural. True epistemic diversity — the kind that prevents two models from converging by default on the same premises — depends on the diversity of providers. It is only between models trained in distinct epistemic cultures that the implicit axioms of one can be contested by the implicit axioms of the other.
Observation of multi-provider sessions empirically confirms the importance of this diversity. Blind spots are not correlated: what a model from a certain provider systematically avoids seeing is not avoided in the same way by a model from another provider. Styles of reasoning genuinely differ: some models tend toward the deductive, others toward the inductive, some toward normative caution, others toward sharp assertion. Cultural, geographical, and ideological sensibilities betray themselves through the selection of examples, the choice of analogies, the implicit hierarchy of considerations to be balanced. This plurality produces what Metamorfon can effectively reveal: not superficial disagreements, but axiological cartographies in which differences translate fundamentally distinct ways of treating a question. Without this plurality, Metamorfon would dialogue with itself under different masks.
What is said here of Metamorfon holds, more broadly, for the collective relationship our societies maintain with artificial intelligence. The plurality of providers is not merely a commercial condition for healthy competition. It is also, and more profoundly, a condition of possibility for informed critique of the models. How are we to evaluate the biases of a model if all dominant models share the same biases? How are we to identify what a provider systematically avoids producing if all available providers avoid the same things? How are we to form a public judgment on the reliability of an artificial intelligence if we lack other intelligences, trained differently, to serve as points of comparison? The diversity of providers is what makes possible the work of contestation internal to the AI ecosystem — work that neither regulators, nor researchers, nor users could perform if a single engineering culture prevailed.
It is in this sense that an AI market dominated by two or three actors would be an impoverishment, and not only in the economic sense of the term. Such a market would suffer not only from a lack of competition, but from a lack of epistemic pluralism. The models produced in such a landscape would be technically well-equipped, but they would share a horizon — a set of presuppositions that the absence of alterity would make impossible to see and contest. Systemic biases would become invisible not because they had been corrected, but because no external point of view would remain to name them. The reduction in the number of actors is not only a market problem, it is a problem of knowledge — and that is a serious reason to wish that alternative providers, including the smallest and the least anglophone, continue to exist and to develop.
That Metamorfon integrates eight providers is therefore not a commercial argument. It is an act that corresponds to a vision: that of an AI that remains contestable because it is plural, and of a dialectical tool that operates only if this plurality is effective. The broad selection of providers available in the application expresses this vision concretely. It is not intended to be exhaustive: other actors will emerge, some will fade, the list will evolve. But it is intended to remain broad in principle, because this breadth constitutes Metamorfon’s raison d’être, and because, more generally, it conditions the epistemic quality of artificial intelligence.