{"id":"openai-gpt-realtime-2-1","nom":"GPT-Realtime-2.1","editeur":"OpenAI","pays":"États-Unis","poids_ouverts":false,"poids_selon":"Artificial Analysis","licence":null,"licence_selon":null,"liens":{"page":"https://developers.openai.com/api/docs/models/gpt-realtime-2.1","essayer":"https://developers.openai.com/api/docs/guides/realtime"},"fiche_auto":false,"page":"https://quelleia.com/voix/modeles/openai-gpt-realtime-2-1/","releve":"2026-11-09","en_bref":"4e sur 12 à l'indice voix à voix d'Artificial Analysis ; 14e sur 38 en raisonnement à l'écoute (Big Bench Audio) ; 9e sur 31 en tenue de la conversation (Full Duplex Bench) (ex æquo à valeur égale) ; 16e sur 34 au délai avant le premier son.","presentation":"Mise à jour de GPT-Realtime-2, le modèle parole à parole d'OpenAI pour les agents vocaux, avec un effort de raisonnement réglable et l'appel d'outils. Elle reconnaît mieux les suites de lettres et de chiffres et gère mieux les silences, le bruit et les interruptions.","sortie":"2026-07-06","langues":{"annoncees":null,"prises_en_charge_selon_open_tts":null,"evaluees":["en"]},"preuves":[{"appuie":["presentation"],"lu":"https://developers.openai.com/api/docs/models/gpt-realtime-2.1.md (« GPT-Realtime-2.1 updates GPT-Realtime-2 with improved alphanumeric recognition, silence and noise handling, and interruption behavior. It supports speech-to-speech interactions with configurable reasoning effort, instruction following, and tool use for complex voice-agent workflows. »), 23/10/2026"},{"appuie":["presentation"],"lu":"https://developers.openai.com/api/docs/models/gpt-realtime-2.1.md (« GPT-Realtime-2.1 supports configurable reasoning effort. Higher reasoning effort can increase latency and output token usage. »), 23/10/2026"},{"appuie":["sortie"],"lu":"https://developers.openai.com/api/docs/changelog.md (« ### Jul 6 Feature · Model: gpt-realtime-2.1 · Model: gpt-realtime-2.1-mini · API: v1/realtime Released [GPT-Realtime-2.1](https://developers.openai.com/api/docs/models/gpt-realtime-2.1), an updated realtime reasoning model »), 23/10/2026"}],"classements":[{"id":"voix-a-voix/indice","nom":"Indice voix à voix d'Artificial Analysis","rang":4,"sur":12,"valeur":79.06,"unite":"points","chiffres":{"cout_par_heure_audio_entree":10.752343244211684,"hebergeur":"OpenAI"},"voisins":["spacexai-grok-voice-2-0","openai-gpt-realtime-2"],"source":{"nom":"Artificial Analysis","mention":"Source : Artificial Analysis","url":"https://artificialanalysis.ai/speech-to-speech","releve":"2026-11-09"}},{"id":"voix-a-voix/raisonnement","nom":"Raisonnement à l'écoute, Big Bench Audio","rang":14,"sur":38,"valeur":95.8,"unite":"%","chiffres":{"cout_par_heure_audio_entree":10.752343244211684,"hebergeur":"OpenAI"},"voisins":["alibaba-qwen-audio-3-0-realtime-flash","spacexai-grok-voice-fast-1-0"],"source":{"nom":"Artificial Analysis","mention":"Source : Artificial Analysis","url":"https://artificialanalysis.ai/speech-to-speech","releve":"2026-11-09"}},{"id":"voix-a-voix/conversation","nom":"Tenue de la conversation, Full Duplex Bench","rang":9,"sur":31,"valeur":95.71,"unite":"%","chiffres":{"cout_par_heure_audio_entree":10.752343244211684,"hebergeur":"OpenAI"},"voisins":["openai-gpt-realtime-1-5","spacexai-grok-voice-2-0"],"source":{"nom":"Artificial Analysis","mention":"Source : Artificial Analysis","url":"https://artificialanalysis.ai/speech-to-speech","releve":"2026-11-09"}},{"id":"voix-a-voix/delai","nom":"Délai avant le premier son","rang":16,"sur":34,"valeur":0.972,"unite":"s","chiffres":{"cout_par_heure_audio_entree":11.306616126202226,"hebergeur":"OpenAI"},"voisins":["google-gemini-3-1-flash-live-preview","openai-gpt-realtime"],"source":{"nom":"Artificial Analysis","mention":"Source : Artificial Analysis","url":"https://artificialanalysis.ai/speech-to-speech","releve":"2026-11-09"}}],"noms_dans_les_sources":{"aa_voix_a_voix":"GPT-Realtime-2.1 High"}}