I-Anthropic S-1, ukwehla kwamanani e-LLM, izingcingo zamathuluzi e-vLLM kanye ne-CNN yesikhumba somdlavuza: izindaba ezine zabathuthukisi ku-Daily Diff eyodwa

Lokho okuhlanganiswa yile vidiyo

• Ifomu le-Anthropic S-1 lamukelwe yi-SEC EDGAR ngo-Okthoba 26, 2026 ngo-09:00 isikhathi saseNew York. Leli ifayela lokuqala: uhla lwamanani, inani lamasheya kanye nezinye izibalo ezintsha kakhulu zingenalutho, kanti inkampani isifake isicelo sokufaka ohlwini amasheya Ekilasi A ku-Nasdaq Global Select Market ngaphansi kwesigama esithi ANT.
• Ithebula lesifingqo le-S-1 libika imali engenayo engu-$16,286 million ngezinyanga eziyisithupha zokuqala zika-2026 (yonke i-2025: $4,585 million), izindleko zokusebenzisa kanye nengqalasizinda ezingu-$8,945 million, ukulahlekelwa kokusebenza okungu-$693 million (umkhawulo wokusebenza -4%), inzuzo yokusebenza elungisiwe (engeyona i-GAAP) engu-$1,618 million efaka emuva isinxephezelo esisekelwe esitokisini esingu-$2,185 million nezinye izinto, kanye nokulahlekelwa okuphelele okungu-$80,748 million. Okungu-$72,388 million wokulahlekelwa yinguquko enanini elifanelekile lamabhondi aguqulekayo, izikweletu eziphuma esivumelwaneni kanye nama-SAFE, okuthi ifayela lisho ukuthi kubangelwe kakhulu ukwanda kwenani lebhizinisi lenkampani.
• I-S-1 iphinde ibike isivumelwano se-Bartz esingu-$1.5 billion (esigunyazwe yinkantolo ngoJulayi 2026), ukheshi kanye nokulingana kokheshi okungu-$112.4 billion ngoJuni 30, 2026, nokuthi abasebenzisi be-Claude Enterprise ngaphandle kwe-Claude Code basebenzise ngaphansi kuka-$75 ngenyanga ngokwesilinganiso ezinyangeni ezintathu kuya ku-Septemba 30, 2026, kanti abasebenzisi abaphezulu abangu-5% be-Claude Enterprise abasebenzise i-Claude Code basebenzise cishe izikhathi ezingu-20 ngaphezu kwalokho.
• I-OpenAI ithi i-GPT-5.6 Luna ibe eshibhile ngo-80% ngoJulayi 30, 2026, futhi ibala i-GPT-6.1 Sol ngo-$2 ukufaka, $10 ukukhipha kanye no-$0.10 ukufaka okugciniwe ngenani lamathokheni ayisigidi, okuyingxenye eyodwa kwezinhlanu zamanani okufaka nawokukhipha ajwayelekile e-GPT-6 Astra. I-Anthropic ithi i-Opus 5.5 ibiza cishe ngo-40% ngaphansi ukuyisebenzisa kune-Opus 5; i-S-1 yayo ithi ukunciphisa kwamanani ngoSepthemba kuhlanganise amanani aphansi nokusebenza kahle okungcono ngenkathi kugcinwa imajini.
• I-Ubicloud (Okthoba 20, 2026) ikale ukunemba kwesikimu socingo lwamathuluzi okungu-92% lwe-Kimi K2.6 ku-vLLM 0.28, okusho ukuthi u-2 kwabangu-25 bacingo lwamathuluzi lwehlulekile, kanye no-100% ngemuva kokuvumela ukuskena okukhawulelwe (ifulegi le-vLLM tool-strict-level, vLLM 0.31) kumamodeli e-Kimi ne-DeepSeek; i-vLLM iyigcina ivaliwe ngokuzenzakalelayo ngoba abasebenzisi babika izinkinga ngamamodeli athile.
• UDkt. Magnus Lynch (i-Guy's Hospital, King's College London; Okthoba 25, 2026) uhlole i-GPT-6 Astra, i-Claude Fable 5.1 kanye ne-Inception v3 CNN elungiswe kahle ezindaweni ezingu-200 eziqinisekiswe yi-biopsy: i-CNN yaqamba ukuxilonga okufanele okungu-125 (62%), ingxoxo ngayinye ngo-42%; i-CNN yaqamba izimila ezingama-13 kwezingama-30 njengokukhetha kokuqala ngokuqhathaniswa no-8 no-11. Ubala amaqiniso yena uqobo, kufaka phakathi izinombolo ezincane kanye nokusebenza kanye ngengxoxo ngayinye; lokhu akusona iseluleko sezokwelapha.

Umbhalo obhaliwe ohunyushwe

Kuhunyushwe kusuka ekulandiseni kokuqala kwesiNgisi. Umsindo namazwibela atholakalayo alawulwa yi-YouTube.

Ukulahlekelwa okungamashumi ayisishiyagalombili ezigidi zamaRandi kuvela kuphi?

0:00 Imali engenayo engamashumi ayisithupha nesithupha ezigidini zamaRandi ezinyangeni eziyisithupha kuzwakala njengenkampani ephrinta imali. I-prospectus ka-Anthropic ngokwayo ibala ukulahlekelwa kwemali engamashumi ayisishiyagalombili ezigidini zamaRandi kulezi zinyanga eziyisithupha. Kule vidiyo, izindaba ezine. Yini ngaphakathi kwe-S-one ka-Anthropic, futhi lokho kulahlekelwa kuvela kuphi? Kungani i-OpenAI ne-Anthropic beyehlisa amanani abo? Ingabe ifulegi elilodwa leseva lingalungisa amamodeli avulekile? Futhi imodeli yezithombe eneminyaka eyishumi ubudala ihlula kanjani izinhlelo zokuxoxa zakamuva ekuboneni

0:28 umdlavuza wesikhumba? Ucwaningo lomdlavuza wesikhumba lwakhiwe ngethuluzi elivela enkampanini esiyindaba yokuqala. Ngizobuyela kulokho ekugcineni. NgoMsombuluko, Okthoba ngama-26, futhi lokhu yi-Daily Diff. I-prospectus yincwadi efakwa yinkampani ngaphambi kokuba ithengise amasheya kumphakathi. Cabanga ngayo njengamanothi okukhishwa ebhizinisi lonke, lapho umugqa ngamunye unezigxivizo zeminwe zommeli. I-Anthropic, inkampani eyenza i-Claude, ifake eyayo yokuqala namuhla.

Yini i-Anthropic eyifakile, futhi yini engenalutho?

0:51 Indaba yokuqala. Ngemuva nje kwehora lesishiyagalolunye ngoMsombuluko ekuseni, isikhathi saseNew York, I-Anthropic yafaka ifomu i-S-one ne-S E C. Iqamba uphawu lwe-Nasdaq A N T. Leli ifayela lokuqala, ngakho intengo yesheya, inani lamasheya kanye nemali engenayo yekota entsha kusalokhu kungenalutho. Imali engenayo ibicishe ibe ngamaRandi ayizigidi ezingamashumi amane nanye ngo-2023 kanti yayicishe ibe ngamabhiliyoni amane nengxenye ngo-2025. Ezinyangeni eziyisithupha zokuqala zonyaka kwakungamabhiliyoni ayishumi nesithupha.

1:19 Khona-ke ukulahlekelwa. Ukulahlekelwa kokusebenza kulezi zinyanga eziyisithupha kwakucishe kube ngamadola ayizigidi eziyikhulu, amaphesenti amane emali engenayo. Ukulahlekelwa okuphelele kwakungamabhiliyoni ayisishiyagalombili, futhi amabhiliyoni angamashumi ayisikhombisa nambili kulokho kungumugqa owodwa, inguquko enanini elifanelekile lamabhondi aguqulekayo. Ngamagama alula, njengoba inkampani iba nenani elikhulu, amabhondi enza aba nenani elikhulu, futhi ama-akhawunti aqopha lokho njengokulahlekelwa. Ifayela liphinde libike inzuzo yokusebenza elungisiwe eyibhiliyoni elilodwa namashumi ayisithupha

1:44 emali, efaka emuva ukukhokhelwa ngamasheya nezinye izinto ezimbalwa. Ifayela elifanayo libonisa nomthethosivivinywa welungelo lobunikazi. Ngo-Septemba 2025 i-Anthropic yavuma ukukhokhela ababhali be-Bartz icala lesikweletu ngemali eyibhiliyoni elilodwa namashumi ayisihlanu emali, elinomugqa walo etafuleni lezindleko. Lokho kuxazululwa kwaphakama endabeni yethu yamabhuku aphehlwe yi-OpenAI ngenyanga eyedlule. Nansi ingxenye yabathuthukisi. Ezinyangeni ezintathu kuya ekupheleni kuka-Septemba, ifayela lithi abasebenzisi be-Claude Enterprise abangazange basebenzise i-Claude Code basebenzise ngaphansi kwamadola angamashumi ayisikhombisa nanhlanu ngenyanga

2:08 ngokwesilinganiso. Abangu-5% abaphezulu abayisebenzise basebenzise cishe izikhathi ezingamashumi amabili kulokho. Ubunikazi buyinqaba. I-trust ikhetha iningi lebhodi ngesigaba sayo samasheya, kanti inkampani ephethwe ngabasunguli ibamba isabelo esisodwa esikhethekile esibiza amaphesenti angamashumi amahlanu nengxenye okhetho. Indaba yesibili. U-Max Woolf, obhala njenge-minimaxir, uhlele zonke izintengo

Kungani i-OpenAI ne-Anthropic beyehlisa amanani abo?

2:28 iguquka kusuka ku-OpenAI naku-Anthropic kusukela ngoJuni, futhi okuthunyelwe kwadlula amaphuzu angamakhulu amathathu namashumi ayisishiyagalombili ku-Hacker News. Okuthunyelwe yi-OpenAI ngokwayo kuthi ngoJulayi 30 i-GPT-5.6 Luna yashibha ngo-80%. Ngo-Septemba 29, i-OpenAI yabala i-GPT-6.1 Sol ngamadola amabili ngenalutho namadola ayishumi ephuma ngenani lamathokheni ayisigidi, ingxenye yesihlanu yenani layo lemodeli ephezulu, i-Astra. I-Anthropic ithi i-Opus 5.5 ibiza cishe ngo-40% ngaphansi ukuyisebenzisa kune-Opus 5, impi yamanani esiyibike ngoSeptemba. Ukusabela kuka-Woolf kwaba ukwehlisa isithunzi.

2:57 Wanciphisa uhlelo lwakhe lwe-Codex lwaba ingxenye yesihlanu yomkhawulo wokusebenzisa, ngoba imodeli efanayo manje iselula kane ngaphezu kwalokho. Uphinde aphawule ukuthi kuma-coding agents, ukufunda kwe-cache kuyiningi lemali yamathokheni. Ifayela le-Anthropic linikeza ingxelo yalo. Lisho ukuthi ukunciphisa kwamanani ngoSeptemba kuhlanganise intengo ephansi nokusebenza kahle okungcono, nokuthi ukusebenza kahle kwekhompyutha kwakwenza ikwazi ukugcina imajini yayo. Ngemuva kwe-Opus 4.5 eshibhile, ithi, ukukhula kwemali engenayo yekota kusuka ku-48% kuya ku-132%

3:23 ngokunye okubangela ukwanda. Incazelo ka-Woolf ngamamodeli avulekile. I-Kimi K 3.1 Pro, ekhishwe ngo-Okthoba 23, ibhalwa ngamanani afanayo angamadola amabili nayishumi njenge-GPT 6.1 Sol. Ngakho-ke phinda ulinganise izindleko zakho ngokusekelwe kumanani akamuva. Emsebenzini we-agent umugqa wokufaka okugciniwe ubaluleke kakhulu, futhi i-OpenAI iwunciphise ngesigamu se-Sol. Indaba yesithathu. I-Ubicloud, inkampani yamafu, ishicilele isixazululo sezincingo zamathuluzi kumamodeli avulekile.

3:49 Isekela i-Kimi K 2.6 ku-vLLM 0.28, amaphesenti angamashumi ayisishiyagalolunye nambili kuphela ezingcingo zamathuluzi ahambelana nesikimu sethuluzi, okusho ukuthi amabili kwangamashumi amabili nanhlanu ahlulekile.

Ingabe ifulegi elilodwa leseva lingalungisa amamodeli avulekile?

4:00 Isixazululo yi-constrained decoding. Iseva ivimba imodeli ekukhetheni noma yiliphi ithokheni elingaphula isikimu se-JSON sethuluzi. Ku-vLLM 0.31 ifulegi libizwa nge-tool strict level, futhi i-Ubicloud ikale u-100% ku-Kimi ne-DeepSeek. I-vLLM iyigcina ivaliwe ngokuzenzakalelayo ngoba abasebenzisi babika izinkinga ngamamodeli athile, njenge-GLM 5.1. Iphuzu le-Ubicloud ukuthi injini nezilungiselelo zayo zishintsha ikhwalithi oyibonayo. Indaba yesine. Udokotela wesikhumba uMagnus Lynch, osebenza e-Guy's Hospital eLondon, unikeze i-GPT 6 Astra kanye ne-Claude Fable 5.1 ukuhlolwa kwezindawo ezingamakhulu amabili

4:33 zesikhumba esinamanxeba, ngasinye njengesithombe esiseduze kanye nesithombe se-dermoscopy. Uphinde walungisa i-Inception v3, inethiwekhi yezithombe i-Google eyishicilele

Imodeli endala ikushaya kanjani izinhlelo zokuxoxa zakamuva?

4:39 ngo-2015, cishe imizuzu engamashumi amabili ku-GPU eyodwa. Kwakulungile ngo-62% wezindawo. Ingxoxo ngayinye yayilungile ngo-42%. Izinhlelo zokuxoxa zazinezinto ezizithandayo. I-GPT 6 Astra ibize izindawo ezingu-75 nge-basal cell carcinoma lapho kukhona eziyi-25. Kwi-melanoma, ebaluleke kakhulu, inethiwekhi encane yaqamba eziyi-13 kweziyi-30 okokuqala, ngokumelene no-8 no-11. ULynch ubala amaqiniso yena uqobo.

5:08 Amakhulu amabili amacala mancane, ingxoxo ngayinye yasebenza kanye, kanti inethiwekhi iyisisekelo esishaywa imikhiqizo elawulwayo. Isebenza kusiphequluli sefoni nge-WebGPU futhi akukho lutho olulayishwayo. Akukho lapha okuyiseluleko sezokwelapha. Futhi intambo kusukela ekuqaleni. ULynch ubhala ukuthi ukuhlaziya kanye nezinhlaka zokuqala zakhiqizwa nge-Claude Code, ithuluzi lokufaka ikhodi le-Anthropic, nokuthi enye yezinhlelo zokuxoxa ezimbili azihlolile nayo engeka-Anthropic. Uma ukhetha ukukufunda lokhu kunokungizwa ngikufunda,

5:37 i-diff ifika ebhokisini lakho leposi njalo ekuseni, mahhala ku-the daily diff dot dev, isixhumanisi esingezansi. Ngakho isinqumo sanamuhla yi-NEEDS REVIEW. Ngizoyibuyekeza ngoba ifayela le-Anthropic linokulahlekelwa ephepheni, inzuzo esilinganisweni esilungisiwe kanye nezinhlobo eziningi ezingenalutho, indaba yentengo incike ingxenye ekufundweni komlobi oyedwa, kanti ukuhlolwa komdlavuza wesikhumba ngamacala angamakhulu amabili aqhutshwa kanye. Futhi lokho yimahluko wanamuhla.

Yini ethola isitembu esithi NEEDS REVIEW?

5:56 NginguNiko ovela ku-Axrisi. Hlanganisa ngokuzibophezela. Isihloko: I-Anthropic S-1, ukunciphisa kwamanani e-LLM, izingcingo zamathuluzi e-vLLM kanye ne-CNN yomdlavuza wesikhumba: izindaba ezine zabathuthukisi ku-Daily Diff eyodwa Isifinyezo: Izindaba ezine zabathuthukisi ezivela ngoMsombuluko owodwa: I-Anthropic yafaka ifomu lomphakathi i-S-1 elibika imali engenayo engu-$16.3 billion kanye nokulahlekelwa kwemali engu-$80.7 billion engxenyeni yokuqala ka-2026, iningi layo liwumugqa wenani elifanelekile kumabhondi aguqulekayo, kanti intengo kanye nenani lamasheya kusalokhu kungenalutho; I-OpenAI ne-Anthropic zehlise amanani e-LLM kanti ifayela linikeza ingxelo ye-Anthropic ngokwayo ngokusika kwayo; I-Ubicloud ibika ukuthi ukuskena okukhawulelwe kuphakamisa ukunemba kwesikimu socingo lwamathuluzi e-Kimi ne-DeepSeek ku-vLLM kusuka ku-92% kuya ku-100%; kanti ukuhlolwa kwezinto ezingu-200 zesikhumba somuntu onesifo esingelapheki kwathola inethiwekhi yezithombe eneminyaka eyishumi ubudala ihlula i-GPT-6 Astra ne-Claude Fable 5.1, ngo-62% kuya ku-42%, ngamaqiniso awabalula yena uqobo. Isimangalo ngasinye sibhekiselwa emthonjeni waso wokuqala, akukho lutho oluyiseluleko sokutshala imali noma sezokwelapha, futhi isinqumo yi-NEEDS REVIEW. Incazelo: Izindaba ezine zabathuthukisi ezivela ngoMsombuluko owodwa: I-Anthropic yafaka ifomu lomphakathi i-S-1 elibika imali engenayo engu-$16.3 billion kanye nokulahlekelwa kwemali engu-$80.7 billion engxenyeni yokuqala ka-2026, iningi layo liwumugqa wenani elifanelekile kumabhondi aguqulekayo, kanti intengo kanye nenani lamasheya kusalokhu kungenalutho; I-OpenAI ne-Anthropic zehlise amanani e-LLM kanti ifayela linikeza ingxelo ye-Anthropic ngokwayo ngokusika kwayo; I-Ubicloud ibika ukuthi ukuskena okukhawulelwe kuphakamisa ukunemba kwesikimu socingo lwamathuluzi e-Kimi ne-DeepSeek ku-vLLM kusuka ku-92% kuya ku-100%; kanti ukuhlolwa kwezinto ezingu-200 zesikhumba somuntu onesifo esingelapheki kwathola inethiwekhi yezithombe eneminyaka eyishumi ubudala ihlula i-GPT-6 Astra ne-Claude Fable 5.1, ngo-62% kuya ku-42%, ngamaqiniso awabalula yena uqobo. Isimangalo ngasinye sibhekiselwa emthonjeni waso wokuqala, akukho lutho oluyiseluleko sokutshala imali noma sezokwelapha, futhi isinqumo yi-NEEDS REVIEW. Takeaway:0: Ifomu le-Anthropic S-1 lamukelwe yi-SEC EDGAR ngo-Okthoba 26, 2026 ngo-09:00 isikhathi saseNew York. Leli ifayela lokuqala: uhla lwamanani, inani lamasheya kanye nezinye izibalo ezintsha kakhulu zingenalutho, kanti inkampani isifake isicelo sokufaka ohlwini amasheya Ekilasi A ku-Nasdaq Global Select Market ngaphansi kwesigama esithi ANT.

Imithombo

1. Anthropic, PBC Form S-1 (registration statement)U.S. Securities and Exchange Commission (EDGAR)
2. EDGAR filing index for accession 0001628280-26-067168U.S. Securities and Exchange Commission (EDGAR)
3. OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?Max Woolf (minimaxir.com)
4. Advancing the price-performance frontier with GPT-5.6OpenAI
5. Introducing GPT-6.1 SolOpenAI
6. Claude Opus 5.5Anthropic
7. Fixing Tool-Call Accuracy in vLLMUbicloud
8. A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosisDr Magnus Lynch (lynchlab.net)
9. Kimi K3.1 ProOpenRouter