Claude Fable 5 from Anthropic Outpaces GPT-5.5 on FrontierMath
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
The Claude Fable 5 model from Anthropic has recently demonstrated its superiority over OpenAI's GPT-5.5 by achieving remarkable scores on the FrontierMath benchmark. With a lead of 13 points, Fable 5 excels in the most challenging problems of this renowned test.
According to Epoch AI, Claude Fable 5 achieves an accuracy of 87% on levels 1 to 3 of FrontierMath and climbs to 88% on level 4, the most complex. This performance marks a significant advancement for Anthropic, whose previous model, Opus 4.5, had scored less than 10% on the same level at the beginning of 2026.
In comparison, OpenAI's GPT-5.5 model shows about 75% accuracy on level 4, although the GPT-5.6 version is already in development. All these models have been evaluated according to Epoch AI's standard scale, which requires maximum reasoning effort.
FrontierMath is recognized as one of the most demanding benchmarks for mathematical reasoning in artificial intelligences. The progress made by the models from Anthropic and OpenAI is not limited to benchmarks: recently, an OpenAI model solved a long-standing Erdős problem, and Claude Mythos also achieved this feat, illustrating the growing capabilities of these technologies.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.