U-Qwen Wethula I-Qwen3.7-Max: Imodeli Ye-ejenti Ebonisayo Ngewindi Lokuqukethwe Le-Token elingu-1M

0 1 6 minutes read

U-Qwen Wethula I-Qwen3.7-Max: Imodeli Ye-ejenti Ebonisayo Ngewindi Lokuqukethwe Le-Token elingu-1M

Amamodeli amaningi we-AI namuhla awaklanyelwe ukukhishwa okuzinzile, okunezinyathelo eziningi. Imisebenzi efana nokwenza amakhulukhulu ezinguquko zekhodi eziphindaphindwayo, noma izingcingo zethuluzi lokuhlanganisa amahora amaningi ngaphandle kokungenelela komuntu, zidinga uhlobo oluhlukile lwemodeli yezakhiwo nokugxila kokuqeqeshwa.

Ithimba lika-Alibaba likaQwen limemezele ngokusemthethweni i-Qwen3.7-Max ku-Alibaba Cloud Summit yango-2026 ngoMeyi 20. Nakuba, izinguqulo ezimbili zokubuka kuqala zochungechunge lwe-Qwen3.7 zivele buthule ebhodini labaphambili le-Arena AI ngaphandle kokukhishwa kwabezindaba futhi kungekho simemezelo esisemthethweni se-API.

Amamodeli Okuhlola Kuqala Amabili Akhishwe Ngasikhathi sinye

U-Alibaba uhlole kuqala amamodeli amabili ngesikhathi esisodwa: Qwen3.7-Max-Preview kanye ne-Qwen3.7-Plus-Preview. Bakleliswe endaweni ye-13 emhlabeni jikelele ngamakhono ombhalo kanye ne-16 emandleni okubona, ngokulandelana, ngokusho kwe-LM Arena.

Ku-Text Arena, i-Qwen3.7-Max-Preview ikleliswe ku-#13 iyonke, ibeka i-Alibaba njengelebhu engu-#6 embhalweni. E-Vision Arena, i-Qwen3.7-Plus-Preview ikleliswe ku-#16 iyonke, ibeka i-Alibaba njengelebhu engu-#5 embonweni. Izinga lemodeli kanye nelebhu yizibalo ezihlukene.

I-Qwen3.7-Plus-Preview ichazwa njengokubuka kuqala kwenguqulo esebenza kahle kakhulu, egxile ekucabangeni nasekukhulumeni okunengqondo, nechungechunge lamathuluzi azovulwa kancane kancane esikhathini esizayo. Iphatha umbono kanye nokokufaka kwe-multimodal. I-Qwen3.7-Max iwuphawu lokubonisana lombhalo kuphela. Lesi sihloko sihlanganisa i-Qwen3.7-Max, njengoba iyimodeli ye-Alibaba eyamenyezelwa ngokusemthethweni ngokufinyelela kwe-API.

Iyini i-Qwen3.7-Max Yakhelwe

Ithimba le-Alibaba Qwen lichaze i-Qwen3.7-Max njengemodeli ye-ejenti ethuthuke kakhulu nephelele kuze kube manje. Imodeli inobunikazi futhi inesisindo esivaliwe. Iyakwazi ukuphatha ukubhala ngekhodi nokulungisa iphutha, ukuzenzela kokuhamba komsebenzi ehhovisi, kanye nemisebenzi yomkhathizwe ethatha amakhulu noma izinkulungwane zezinyathelo.

Imodi Yokucabanga Enwetshiwe

I-Qwen3.7-Max iyimodeli yokucabanga. Imodeli ikhiqiza uchungechunge lomcabango kuqala – ukulandelana kwangaphakathi kwezinyathelo lapho ihlela khona, ihlola umsebenzi wayo, futhi ilungise inkambo ngaphambi kokuzibophezela empendulweni yokugcina. Ezingosini ezifana ne-Qwen Chat, lokhu kubonakala njengemodi 'Yokucabanga' ongayivula ukuze ubone umkhondo wokucabanga wemodeli.

Amamodeli okucabanga akhiqiza amathokheni okukhiphayo kakhulu kunokuqedwa okujwayelekile. Lapho i-Artificial Analysis yenza ukuhlolwa kwe-Intelligence Index, i-Qwen3.7-Max ikhiqize amathokheni angaba yizigidi ezingu-97, uma kuqhathaniswa nesilinganiso sezigidi ezingu-24 zamamodeli akulelo bhentshimakhi. Ngemisebenzi emifushane noma elula, lokhu kungaphezulu kwengeza ukubambezeleka ngaphandle kokuthuthukisa ikhwalithi yokuphumayo. Ngokuhlela okunezinyathelo eziningi, ukwenza kabusha ikhodi, noma amaketanga e-ejenti amade, imodi yokucabanga eyandisiwe yilapho amandla emodeli esebenza khona.

Iwindi Lokuqukethwe

Imodeli inewindi lomongo wethokheni elingu-1M, lisuka ku-256K ku-Qwen3.6 Max Preview. Isekela okokufaka nokukhishwayo kombhalo kuphela. Intengo ayikamenyezelwa. I-Qwen3.6 Max Preview inenani elingu-$1.30/$7.80 ngesigidi samathokheni okokufaka/okukhiphayo ku-Alibaba Cloud.

Iwindi lomongo wamathokheni ayisigidi lingabamba inqolobane egcwele yekhodi yosayizi omaphakathi noma inqwaba yamadokhumenti esicelweni esisodwa. Amamodeli avame ukucabanga ngokungathembeki njengoba iwindi lokuqukethwe ligcwala. Ukuhlola okuzimele kokuqukethwe okude kwe-Qwen3.7-Max akukakatholakali.

Imiphumela yeBenchmark

I-Qwen3.7-Max ithole u-56.6 ku-Artificial Analysis Intelligence Index, okuyibeka endaweni yesihlanu isiyonke. Lokho kumelela ukuzuza kwamaphoyinti angu-4.8 ngaphezu kweyandulela i-Qwen3.6 Max Preview (51.8), futhi ikubeka ngaphambi kwe-Gemini 3.5 Flash ye-Google (55.3). I-GPT-5.5 (60.2), Claude Opus 4.7 (57.3), kanye ne-Gemini 3.1 Pro Preview (57.2) zisahola phambili ekulinganisweni okuphelele.

I-Intelligence Index v4.0 ihlanganisa ukuhlola okuyishumi, okuhlanganisa i-GDPval-AA, Terminal-Bench Hard, SciCode, AA-Omniscience, Humanity's Last Exam, kanye ne-GPQA Diamond.

Ukuthuthukiswa ngaphezu kwe-Qwen3.6 Max Preview akufani. Iningi lezinzuzo ze-Index zigxile ekucabangeni kwesayensi, emandleni okusebenza, nasekubhaleni amakhodi. I-CritPt ikhuphuke ngamaphesenti angu-9.7 (isuka ku-3.7% yaya ku-13.4%), i-Humanity's Last Exam igxume ngamaphuzu angu-9.2 (isuka ku-28.9% yaya ku-38.1%), kanti i-Terminal-Bench Hard ikhuphuke ngamaphuzu angu-6.9 (isuka ku-43.9% yaya ku-50.8%). I-GDPval-AA yengeze amaphuzu angu-42 Elo (kusuka ku-1504 kuya ku-1546). Izikolo kwamanye amabhentshimakhi ziphansi kakhulu uma kuqhathaniswa ne-Qwen3.6 Max Preview.

Omunye umphumela kuyi-Index udinga ukufundwa ngokucophelela. Ku-AA-Omniscience, ukunemba okuluhlaza kwe-Qwen3.7-Max empeleni kwehle ngamaphesenti angu-7.6 (kusuka ku-37.7% kuya ku-30.1%), kuyilapho izinga layo lokungaboni kahle lehle ngamaphuzu angu-21.3 (kusuka ku-44.2% kuya ku-22.9%). Imodeli ikhetha ukuthi “angazi” kaningi kunokukhumbula amaqiniso amaningi. Izinga layo lokuzama lehle lisuka ku-67.3% laya ku-48.0%, eliphansi kakhulu phakathi kwamamodeli asemngceleni uma kuqhathaniswa. Ibhentshimakhi ye-AA-Omniscience iklomelisa izimpendulo ezifanele futhi ijezisa imibono engemihle kodwa ayinaso isijeziso sokwenqaba ukuphendula. Ezimweni zokusetshenziswa ezincike ekukhunjulweni kweqiniso elibanzi, lokhu kuwumkhawulo ozwakalayo wokuhlola umsebenzi wakho.

Ku-Text Arena, i-Qwen3.7-Max-Preview ikleliswe ku-#13 iyonke ngesikolo se-Elo esingu-1,475. Amazinga esigaba afaka #7 ku-Math, #9 ku-Expert Prompts, #9 ku-Software ne-IT, kanye ne-#10 ku-Coding.

Zonke izinombolo zebhentshimakhi zingezandulela. Imodeli iphethe imodi 'Yokubuka kuqala', okubonisa ukuthi i-Alibaba ikuthatha njengesakhiwo sangaphambi kwesikhathi.

Ukusebenza Kwe-Agent – Ukuhlolwa Kwangaphakathi

Ohlolweni lwangaphakathi lwe-Alibaba kuplathifomu ye-chip entsha, imodeli yenze ngokuzenzakalelayo amakholi amathuluzi angaphezu kuka-1,000 kanye nokuguqulwa kwekhodi ephindaphindwayo ukuze kulungiselelwe ukhiye wokhiye. U-Alibaba uthe le nqubo ithuthukise isivinini sokunquma cishe ngo-10x uma iqhathaniswa nenguqulo yangaphambilini.

Isichazi Esibonakalayo sikaMarktechpost

Isilayidi 1 kwezi-6

Yini i-Qwen3.7-Max?

Imodeli yokucabanga yobunikazi evela ku-Alibaba, edizayinelwe imisebenzi ye-ejenti yendawo ende, ukukhiqizwa kwekhodi, kanye nokuzenzakalela kwezinyathelo eziningi.

Iwindi Lokuqukethwe

Amathokheni ayizigidi ezingu-1 – anele ukufaka ikhosombe lekhodi eliphakathi nendawo esicelweni esisodwa.

Imodeli Yokubonisana

Isebenzisa uchungechunge lokucabanga (imodi yokucabanga eyandisiwe) ngaphambi kokukhiqiza impendulo yokugcina.

Okokufaka / Okukhiphayo

Thumela umbhalo, khipha umbhalo. Akukho okokufaka kwesithombe okusekelwe kule modeli.

I-API String

Sebenzisa qwen3.7-ubukhulu lapho ushaya nge-Alibaba Cloud Model Studio.

I-Apache-compatible API
I-OpenAI ne-Anthropic spec
Buka kuqala — azikho izisindo ezivuliwe okwamanje

Isilayidi 2 kwezi-6

Ukuqala Okusheshayo: Isixhumi esibonakalayo sengxoxo

Indlela eshesha kakhulu yokuhlola i-Qwen3.7-Max ngaphandle kokhiye we-API noma ukusetha okudingekayo.

1

Iya kuQwen Chat

Zulazulela ku chat.qwen.ai futhi udale i-akhawunti yamahhala.
2

Khetha imodeli

Kokudonsela phansi kwesikhethi semodeli, khetha Qwen3.7-Ubukhulu. Kungase kubonakale njenge Qwen3.7-Max-Preview ngesikhathi sokubuka kuqala.
3

Nika amandla Imodi yokucabanga

Guqula uvule Imodi yokucabanga kusixhumi esibonakalayo sengxoxo. Lokhu kwenza kusebenze ukucabangana okulandelanayo futhi kubonisa umkhondo wangaphakathi wokucabanga wemodeli ngaphambi kwempendulo yokugcina.
4

Thumela umyalezo wakho

Thayipha umbuzo wakho. Ukuze uthole imiphumela engcono kakhulu emisebenzini eyinkimbinkimbi, cacisa mayelana nezinyathelo, imikhawulo, nefomethi elindelekile yokuphumayo.

💡

Sebenzisa ukwaziswa kwakho komhlaba wangempela okunzima kakhulu uma uhlola. Izinkinga zezibalo zezinyathelo eziningi, izicelo eziyinkimbinkimbi zokuphinda zenziwe kabusha, kanye nemibuzo yochwepheshe engacacile yembula okuningi ngekhwalithi yemodeli kunemiyalelo elula.

Isilayidi 3 kwezi-6

Ukufinyelela kwe-API

I-Qwen3.7-Max ihambisana nakho kokubili okucacisiwe kwe-OpenAI ne-Anthropic API. Ungayixhuma kumapayipi akhona ngezinguquko ezincane.

Ucingo lwePython oluhambisana ne-OpenAI

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_DASHSCOPE_API_KEY",
    base_url="
)

response = client.chat.completions.create(
    model="qwen3.7-max",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user",   "content": "Explain chain-of-thought reasoning."}
    ]
)

print(response.choices[0].message.content)

ℹ️

Thola ukhiye wakho we-API kusuka I-Alibaba Cloud Model Studio (I-DashScope). I-URL eyisisekelo yokufinyelela kwamanye amazwe ithi dashscope-intl.aliyuncs.com.

⚠️

Amanani awakamenyezelwa we-Qwen3.7-Max. Ukuze uthole inkomba, I-Qwen3.6 Max Preview inenani elingu-$1.30 / $7.80 ngesigidi samathokheni okokufaka/okukhiphayo.

Isilayidi 4 kwezi-6

Ukuqonda Imodi Yokucabanga

Imodi yokucabanga iyisendlalelo sokucabanga semodeli. Inquma ukuthi imodeli ibhekana kanjani nenkinga ngaphambi kokukhiqiza impendulo.

Isetshenziswa nini

Ukwenziwa kabusha kwekhodi yezinyathelo eziningi, ubufakazi bezibalo obuyinkimbinkimbi, amaketanga emisebenzi ama-ejenti amade, nezinkinga ezingacacile ezidinga ukuhlelwa kwesinyathelo nesinyathelo.

Nini ukweqa

Ukubhala kabusha okufushane, izigaba ezilula, ukubheka ngokushesha, noma imisebenzi lapho ukubambezeleka kanye nezindleko zamathokheni kudingeka kuncishiswe.

I-API: Nika amandla ukucabanga nge-extra_body

response = client.chat.completions.create(
    model="qwen3.7-max",
    messages=[{"role":"user","content":"Your prompt here"}],
    extra_body={"enable_thinking": True}
)

💡

I-Qwen3.7-Max ikhiqize amathokheni angu-97M kumabhentshimakhi okuhlaziya okwenziwayo, uma kuqhathaniswa nesilinganiso esingu-24M samamodeli aqhathanisekayo. Ithokheni yokucabanga ngayinye yengeza ekubambezelekeni nasezindlekweni – sebenzisa imodi yokucabanga ngokukhetha.

Isilayidi 5 kwezi-6

Imisebenzi ye-Agentic kanye ne-Long-Horizon

I-Qwen3.7-Max yakhelwe ukusebenzisa amalophu emisebenzi amade, azimele. Ekuhloleni kwangaphakathi kwe-Alibaba, yenze izingcingo zamathuluzi ezingu-1,000+ futhi yazinza ngokuzenzakalelayo amahora angafinyelela kwangu-35.

1

Chaza amathuluzi ngokucacile

Dlula izincazelo zamathuluzi ku-OpenAI evamile amathuluzi ipharamitha. Imodeli isekela ukushaya komsebenzi kanye nesicelo sethuluzi eliphindaphindayo ngokomdabu.
2

Sebenzisa iwindi lokuqukethwe elingu-1M ngenhloso

Dlula umlando ogcwele womsebenzi, okuphumayo kwethuluzi langaphambili, kanye nesimo sekhodi sibe umongo. Sika ngamandla lapho umongo ogcwele ungadingeki — yonke ithokheni iyakhokhiswa.
3

Khomba impendulo yokugcina ngokugomela

Okuphumayo kokucabanga yinde futhi kuyaguquguquka kunokuqedwa okujwayelekile. Lapho ubhala izivivinyo, gcizelela impendulo yokugcina, hhayi amagama aqondile omkhondo wokucabanga.
4

Amacala okusebenzisa kahle

Ukulungiselelwa kwe-Kernel, amalophu okulungisa amakhodi, ukuzenzekelayo kokugeleza komsebenzi kwehhovisi, kanye nemigqa yedatha enezinyathelo eziningi enokuqinisekiswa okuphindaphindiwe.

⚠️

Izibalo zekholi zamathuluzi zamahora angu-35 nezingu-1,000+ zivela ekuhlolweni kwangaphakathi kwe-Alibaba kuphela. Akukho ukuqinisekiswa okuzimele okukhona kwalezi zimangalo ezithile.

Isilayidi 6 kwezi-6

Imikhawulo Eyaziwayo

Ukuqonda le mikhawulo ngaphambi kokuhlanganisa kuzokonga isikhathi sokulungisa iphutha futhi kukusize usethe okulindelekile okulungile.

Akukho okokufaka kwesithombe

I-Qwen3.7-Max iwumbhalo kuphela. Ngemisebenzi ehlukahlukene, sebenzisa i-Qwen3.7-Plus-Preview esikhundleni, esekela okokufaka kokubona.

AA-Omniscience abstention

Kubhentshimakhi ye-AA-Omniscience, izinga lokuzama lemodeli lehlile lisuka ku-67.3% laya ku-48.0%. Igodla okuningi futhi ibona izinto ezingekho ngaphansi – kodwa ukukhumbula kwayo okungelona iqiniso nakho kwehlile. Hlola ngokucophelela imisebenzi yokukhumbula ulwazi.

Buka kuqala isimo

Imodeli okwamanje iphethe – Buka kuqala isijobelelo. Izikolo zebhentshimakhi, ukuziphatha, kanye nentengo kungashintsha ngaphambi kokukhishwa okuzinzile. Ayikho inguqulo yesisindo esivulekile etholakalayo kusukela ngoMeyi 2026.

Ukuthembeka kokuqukethwe okude

Iwindi lokuqukethwe kwethokheni elingu-1M liwusilingi, akusona isiqinisekiso. Ukuhlola okuzimele kokuqukethwe okude kwe-Qwen3.7-Max akukakatholakali. Qinisekisa ikhwalithi yokubuyisa emthwalweni wakho womsebenzi othile.

ℹ️

Ukuze uthole izibuyekezo zakamuva zemodeli, hlola ibhulogi esemthethweni ye-Qwen kokuthi qwen.ai/blog kanye namadokhumenti e-Alibaba Cloud Model Studio.

Okubalulekile Okuthathwayo:

I-Alibaba ikhiphe amamodeli amabili wokubuka kuqala we-Qwen3.7: Ubukhulu (umbhalo/ukubonisana) kanye ne-Plus (i-multimodal).
U-Qwen3.7-Max uthole u-56.6 ku-Artificial Analysis Intelligence Index, ekleliswe ku-#5 iyonke — inzuzo enamaphuzu angu-4.8 ngaphezu kokubuka kuqala kwe-Qwen3.6 Max.
Iwindi lokuqukethwe kwethokheni elingu-1M liphinda kabili umkhawulo ongu-256K ukusuka ku-Qwen3.6 Max Preview; umbhalo kuphela, akukho okokufaka kwesithombe.
Ku-AA-Omniscience, ukunemba okuluhlaza kwehlile ngenkathi ukuvinjelwa kukhuphuka – ukuhlolwa okufanele kuhlolwe amacala okusebenzisa ukukhumbula ulwazi.
Imodeli isekele izingcingo zamathuluzi eziyi-1,000+ kanye nokusebenza okuzenzakalelayo kwamahora angama-35 ekuhloleni kwangaphakathi kwe-Alibaba kuphela; akukho ukuqinisekiswa okuzimele okwamanje.

Hlola Imininingwane yobuchwepheshe. futhi Amadokhumenti. Futhi, zizwe ukhululekile ukusilandela Twitter futhi ungakhohlwa ukujoyina wethu 150k+ ML SubReddit futhi Bhalisela ku Iphephandaba lethu. Linda! ukutelegram? manje ungasijoyina kuthelegramu futhi.

Udinga ukusebenzisana nathi ekuthuthukiseni i-GitHub Repo yakho NOMA Ikhasi Lobuso Lokugona NOMA Ukukhishwa Komkhiqizo NOMA I-Webinar njll.? Xhumana nathi

Source link

nimda 1 hour ago

0 1 6 minutes read