U-Anthropic umemezele ngomhla ka-10 ngesikhathi sendawo ukuthi imodeli yakhe yobuhlakani bokwenziwa i-Claude Sonnet 5, eyethulwe ekupheleni kukaJuni kulo nyaka, izogcina unomphela intengo ekhethekile yokuphromotha yangesikhathi yethulwa. Intengo yokuqala ebihlelelwe ukusebenza ngomhla ka-1 Septhemba ikhanseliwe.
Intengo yamanje ye-Claude Sonnet 5 ingamadola angu-2 ngokufakwa kwama-Token ayisigidi kanye namadola angu-10 ngokukhishwa kwama-Token ayisigidi. “Intengo yokuqala” ebihlelelwe ukusetshenziswa ngaphambilini ibingamadola angu-3 / 15, okusho ukwenyuka ngo-50%.
I-Claude Sonnet 5 yakhululwa ngomhla ka-30 Juni 2026, futhi iyimodeli ye-Sonnet enamandla kakhulu ngokwezinga lokusebenza njengomenzeli u-Anthropic ake wayikhiqiza kuze kube manje. Le modeli inewindi lomongo elinama-Token ayisigidi, ikhiphe ama-Token afinyelela ku-128K, kanti ulwazi lwayo lugcina ngoJanuwari 2026. Uma iqhathaniswa ne-Sonnet 4.6 yangaphambilini, yenze intuthuko enkulu ekucabangeni, ekusetshenzisweni kwamathuluzi, ekuhleleni izinhlelo nasekusebenzeni ngolwazi. Ibonise ukusebenza okuhle ezivivinyweni ezifana ne-BrowseComp yokusesha kwabenzeli kanye ne-OSWorld-Verified yokusetshenziswa kwamakhompyutha, kanti ukusebenza kweminye imisebenzi kusondele ku-Opus 4.8 esezingeni eliphezulu. Ekuhlolweni kokuhlela izinhlelo kwe-SWE-bench Verified, i-Sonnet 5 ithole amaphuzu angu-85.20%.
Kumakhasimende ezinkampani nabathuthukisi be-AI esebenza njengomenzeli, ukuzinza kwesikhathi eside kwentengo ye-API sekuba yinto ebalulekile ekusetshenzisweni kokuhamba komsebenzi kwe-AI. Ezimo ezifana nokuthuthukiswa kwekhodi, ucwaningo kanye nokuzenzakalela kweziphequluli, imisebenzi yabenzi imvamisa idinga ukubizwa kwemodeli kaningi, ngakho izindleko zama-Token ziqongelela ngokushesha njengoba umsebenzi uba yinkimbinkimbi. Ukukhansela kuka-Anthropic ukwenyusa intengo kunikeza izinkampani esezakhe ukuhamba komsebenzi okusekelwe ku-Sonnet 5 isibikezelo esizinzile sezindleko zesikhathi eside. Lokhu kubhekwa njengengxenye yesu lika-Anthropic lokusuka emncintiswaneni ogxile ekusebenzeni kuphela iye emncintiswaneni ogxile emandleni aphelele ezindleko kumamodeli e-AI.