Home / Choosing your model and your tool

The model landscape, one family of predictors

Claude, GPT, Gemini and open-weight models are all next-word predictors trained differently, and a prompt template written for one carries across to the others with barely any change.

Claude, GPT and Gemini rest on the same basic training principle. Anthropic describes the initial learning phase of Claude this way: the model learns to predict the next word from all the text that precedes it in the document. The models offered by other major providers, and the open-weight models that anyone can install on their own hardware, share this same general architecture, trained on corpora and with settings that differ from one provider to the next.

What carries over from one assistant to another

This kinship explains why a prompt-writing technique learned on Claude often works almost as it stands elsewhere. A template that sets a role, an expected output format and a sample response is interpreted in a comparable way, because each model completes the text that follows while taking that context into account, whichever provider trained it.

You write minutes for association meetings.
Answer in three numbered points, each a complete sentence.
Here are the raw notes to summarise: notes

This template pastes into another conversational assistant with almost no change: the role, the three-point format and the expected example remain understandable, because they describe an instruction rather than a technical command specific to Claude. What is left is to compare the answers obtained, which choosing the right model tier for the task helps do methodically rather than by chance.

What actually sets providers apart

The level of reasoning brought to bear on an answer also varies within a single family, not only between providers. A model built for speed answers faster but digs less deeply into a complex problem than a model built for depth, a distinction found at Claude just as at its competitors, under names that differ from one to the next.

Shared architecture says nothing about the practical differences, which are real. The size of the context window, the way each provider handles the data sent to it, and the ecosystem of tools around the model, connectors, memory and scheduled tasks for Claude, change from one provider to the next and evolve over time. The figure below compares these criteria without fixing an exact figure, a figure for a context window or a tier name goes stale faster than the principle it illustrates.

Checking these differences at the moment of choosing remains more reliable than relying on memory. A prompt that works well on one provider can give a shorter, longer or less structured answer on another, even when its basic structure carried across without notable change.