Kwenzekeni
I-GIGAZINE ibika ukuthi i-Cognition, inkampani elandela u-Devin, ikhiphe i-Fusion, ihhanisi eklanyelwe ukuthuthukisa indlela amamodeli e-AI athuthukile ahlela ngayo, akhiphe, abuyekeze, futhi alulame emisebenzini emisiwe. Uhlelo lubhanqa imodeli ephezulu yokuhlela nokubuyekeza ngemodeli engabizi kakhulu ukuze isetshenziswe, kuyilapho isebenzisa ama-ejenti amabili ngokuhambisana nezimo ezihlukene namathuluzi.
I-GIGAZINE ibika ukuthi i-Cognition ithuthukise futhi yakhulula i-Fusion njengehhanisi le-OpenAI's GPT-6 Astra kanye ne-Anthropic's Claude Fable. I-athikili ichaza ihhanisi njengesendlalelo esinikeza amamodeli anemiyalo yokusebenza, imithetho yokusetshenziswa kwamathuluzi, namagatsha anemibandela, okungaba nomthelela ekutheni umenzeli uyawuqedela umsebenzi noma abambeke ku-loop.
Ngokusho kwe-GIGAZINE, i-Fusion ihlanganisa imodeli yesimanje yokuhlela nokubuyekezwa ngemodeli engabizi kakhulu yokusebenza. I-Cognition incose i-GPT-6 Astra ne-Claude Fable njengamamodeli athuthukile, futhi yahlonza imodeli yekhodi ye-Cognition ye-SWE-2 njengenketho eshibhile yokwenza. Isihloko sithi inhlanganisela yenze kahle kakhulu nge-Claude Fable 5.1.
I-GIGAZINE ibika amaphuzu ebhentshimakhi angu-62.2 we-Claude Fable 5.1 ne-Claude Code, uma kuqhathaniswa no-61.7 we-Claude Fable 5.1 kuhlanganiswe nemodeli ebiza kancane kanye ne-Fusion. Inhlanganisela yokugcina kubikwe njengeshibhe ngo-36%. Ku-GPT-6 Astra, i-athikili ithi i-Fusion izuze ukusebenza okulingana ne-Codex ngenkathi yehlisa izindleko ngo-39%.
I-athikili ibeka ukulinganisa ezinkampanini ze-AI zokuhlaziya i-Artificial Analysis kanye ne-Vals AI. Ithi i-Fusion isebenzisa ama-ejenti amabili ngokufana, ngalinye linomongo ozimele namathuluzi, futhi lishintshanisa kuphela izifinyezo, imiphumela, nempendulo kunengxoxo egcwele. UGIGAZINE uthi le ndlela isebenzisa kakhulu ukugcinwa kwesikhashana ngokushesha. Umthombo awunikezi indlela ephelele yebhentshimakhi, amanani aphelele, noma imigomo yokutholakala.
Imininingwane yomthombo: gigazine.net ↗
Kungani kubalulekile
Uma imiphumela ebikiwe ibambekile, i-Fusion inganciphisa izindleko zokusebenzisa ama-agent e-frontier AI ngaphandle kokulahlekelwa okuqhathanisekayo ekusebenzeni kwekhodi. Lokho kubalulekile konjiniyela nezinhlangano ezisebenzisa amasistimu okusebenzisa amathuluzi esikalini, lapho imodeli yokucabanga kanye nemizamo ephindaphindiwe ye-ejenti ingaba yizindleko ezinkulu. Imiphumela ibikwe yi-GIGAZINE futhi ihlolwe i-Artificial Analysis kanye ne-Vals AI, kodwa umthombo awunikezi indlela eyanele yokuhlola ngokuzimele ukuqhathanisa.
Imiphumela ebikiwe iphakamisa ukuthi ukwakheka kwe-ejenti kungathinta kakhulu umnotho wamamodeli wemingcele. Isistimu ethumela ukusebenza kumodeli eshibhile kuyilapho igcina imodeli eqinile yokuhlela nokubuyekeza ingase yehlise izindleko zama-ejenti esofthiwe ngaphandle kokudinga abasebenzisi ukuthi balahle amamodeli wamandla aphezulu.
Ukwehliswa kwezindleko kubaluleke kakhulu kuma-ejenti okubhala amakhodi, angenza amakholi amaningi amathuluzi, aphinde imizamo ehlulekile, noma agcine izimo ezinde. Uma kukhiqizwa kabusha ngokuzimela, ukwehliswa okubikwe ngo-36% no-39% kungase kube nomthelela endleleni izinkampani eziklama ngayo ukugeleza komsebenzi komenzeli wokukhiqiza futhi zikhethe phakathi kwamahhanisi obunikazi.
Ubufakazi buhlala bunomkhawulo embikweni onikeziwe. I-GIGAZINE inikeza amaphuzu akhethiwe kanye namaphesenti okonga kodwa hhayi isethi yomsebenzi, ukubala okwenziwayo, ukuqagela kwentengo, ukubambezeleka, amazinga okuhluleka, noma izimo zokuqhathanisa. Ukulingana okufunwayo kwe-Codex kanye nezinzuzo ezishiwo ngokusebenza kahle aziqinisekiswanga ngokuzimela lapha.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Ongakubuka ngokulandelayo
Imibuzo ebalulekile iwukuthi ubani ongafinyelela i-Fusion, imaphi amamodeli namathuluzi ewasekelayo, noma ngabe iyatholakala ngokuvamile, nokuthi izindleko zayo zibalwa kanjani. Ukubika okwengeziwe kufanele kucacise imisebenzi yebhentshimakhi, izisekelo, osayizi besampula, ukuqagela kwe-cache esheshayo, kanye nokuthi imiphumela idlulisa yini ngale kokubhalwa kwekhodi. Umthombo uphinde usebenzise kokubili i-“SWE-2” kanye ne-“SWR-2” kumodeli ebiza kancane, ukungqubuzana okufanele kuxazululwe.
Izimo zokufinyelela azibhalwanga emthonjeni. Akwaziwa ukuthi i-Fusion iyadawunilodeka yini esidlangalaleni, itholakala ngemikhiqizo ye-Cognition, ikhawulelwe kubasebenzisi abakhethiwe, noma inikezwa ngolunye uhlelo lwezentengiso. Intengo kanye nanoma yikuphi ukubhaliswa kwamamodeli adingekayo akwaziwa.
Ukubika kokulandelela kufanele kubonise ukuthi i-Fusion iyawasekela yini amamodeli angaphezu kwe-GPT-6 Astra kanye nenganekwane ye-Claude, ukuthi abasebenzisi bangakwazi yini ukuhlinzeka ngamathuluzi abo kanye nokwaziswa, nokuthi kungakanani ukonga kuncike ekugcinweni kwesikhashana okusheshayo noma emananini athile omhlinzeki.
Ukuqanjwa kwebhentshimakhi kuqukethe ukungahambisani kweleveli yomthombo: i-athikili ikhomba imodeli eshibhile yekhodi njenge-SWE-2 kodwa kamuva ibhekisele ku-“SWR-2.” Lokho kuqokwa kufanele kuqinisekiswe ngaphambi kokuphatha ukuqhathanisa njengesimangalo esinembile somkhiqizo.
Ukuhlola okuzimele kufanele kuhlole amazinga empumelelo yokubhala ikhodi, ukubambezeleka, ukwethembeka, ukuphatha umongo, nokusebenza kwemisebenzi ngaphandle kwebhentshimakhi ebikiwe. Ama-Parallel agents angase futhi andise inkimbinkimbi ye-orchestration noma umsebenzi wamathuluzi aphelele nanoma izindleko zemodeli zehla.