Gemini 4 Post-Training, First Version Expected by Year-End
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Google DeepMind indicates that Gemini 4 has already passed the post-training stage. A first version is expected well before the end of the year, but no firm date, technical details, or access conditions have been provided. The announcement does not take the form of a complete official launch.
No benchmarks, no pricing, and undefined access
To date, no public information allows for an assessment of the progress made since Gemini 3. Google does not reveal the architecture, the number of parameters, or the size of the context window. Supported modalities, scores obtained during benchmarks, execution speed, pricing, and access conditions for the API have not been disclosed. No specific launch date has been announced. The company does not specify whether the initial version will be open to the general public, reserved for developers via an API, or restricted to certain services. The term "release" does not necessarily imply immediate global availability. In the absence of these elements, it is impossible to determine whether Gemini 4 will create a technical gap or simply bring the range up to par with its rivals.
Tight timeline but no firm appointment
Google hopes to offer a first version well before the end of the year, then accelerate iterations. The group presents a tight timeline, without committing to a specific date. The mentioned perspective remains compatible with an arrival well before December 2026. The information is not delivered as part of a complete official announcement, leaving many unknowns regarding the concrete modalities.
Post-training confirmed and focus on rapid updates
Koray Kavukcuoglu, who has just been appointed to lead Google DeepMind, states that Gemini 4 is already in the post-training phase. This stage, which follows the initial training, aims to refine the model's behavior before it becomes available. According to the executive, the performance observed internally encourages DeepMind to accelerate the process, promising regular updates after the initial launch. Despite this confirmation of progress, Google currently provides no technical details to evaluate the model.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.