API GLM Kore-ārai: te uara pai rawa ia tohu
Te tūāpapa GLM kaihautū a Z.ai, e mahi ana kāore he paparanga whakakāhore i runga i tā mātou ake tautau GPU — horopaki 262K, karanga taputapu, me tētahi utu whakaranu iti iho i te haurua o te utu ā-kamupene tonu o taua whakapapa ake.
Whakarāpopoto
GLM-5.2-3BIT-REAP ko te tauira i te Kūwaha Shannon AI e tūtohutia taunoatia ana e mātou. Ka whakahaerea e ia he tūāpapa Whakaranu-o-Ngā-Mātanga korekore 744-piriona tawhā nō te whakapapa GLM — kua whakangungua mārōrō mō te whakamahi taputapu kaihautū mai i te GLM-4.5 — kua kōpakina ki ngā taumaha 3-bit me te tapahi mātanga REAP kia uru ai, kia noho tere tonu ai. Ka riro i a koe he horopaki 262,144 tohu, 16,384 tohu whakaputa, te karanga taputapu, te whakaaroaro me te reasoning_effort tāritari, me te putanga hanganga JSON Schema, mō te $0.73 / $2.34 ia miriona tohu. Kāore he paparanga whakakāhore, kāore he tātari whakaputa. He kuputuhi anake — kāore he tirohanga.
He tinihanga te nuinga o ngā whakatauritenga "tauira utu-iti rawa", nā te mea he tauira iti te kōwhiringa utu-iti e mau ana i te panui a tētahi tauira nui. Ehara tērā i te mea e puta ana i konei. Ko GLM-5.2 tētahi o ngā tūāpapa tuwhera-taumaha nunui rawa kua tukuna e te tangata, ā, i te $0.73 tāuru / $2.34 whakaputa, kotahi tonu te taumata i runga ake i te papa o tā mātou rārangi, engari e kawe ana i tētahi tūāpapa he maha kē ake ōna whā i tā taua papa e hoko ai i te nuinga o te wā. Ko taua āputa — tauira nui, utu tata-papa — te take e noho ai tēnei whārangi.
01Nō whea a GLM: Tsinghua, te whakakī-tūwātawata, me te huri kaha ki ngā kaihautū
Ehara a GLM i te uri Llama kua tapaina anō. He mea tērā o ngā whānau tauira torutoru e whai ana i tāna ake whāinga whakangungu-tōmua, kua whakahoungia mō ngā tau e rima e taua rōpū rangahau kotahi tonu. Ko te ingoa he pou mō te General Language Model, ā, ko te pepa whakatū — GLM: General Language Model Pretraining with Autoregressive Blank Infilling, ACL 2022 — i whakatakoto i tētahi kōwhiringa tuatoru i te taha o te hoahanga reo-taupokina a BERT me te autoregression matau-mā-katau pūmau a GPT. Ka whakatūwātawata a GLM i ngā wāhanga hono, ka whakaputa anō mā te autoregression i tētahi raupapa kua whakaranua, me ngā whakawaehere tūranga rua-ahanoa e aro ana ki te tūranga o tētahi tohu i roto i te tuhinga kua hē me tōna tūranga i roto i te wāhanga e whakakīa ana: kotahi te hoahanga e whakahaere ana i te māramatanga me te whakaputa, e ōrite ana te pai o ngā mea e rua.
Ko te rōpū i muri i tēnei ko te Knowledge Engineering Group o Tsinghua University, nāna i puta ai te kamupene e mōhiotia nei ko Z.ai (i mua ko Zhipu AI). Ko tāna hua matua tuatahi ko GLM-130B i te Whiringa-ā-nuku 2022 — he tauira kiato reo-rua Ingarihi/Hainamana 130-piriona-tawhā, he mea tuku tūmatanui ngā taumaha me te waehere i te wā i whakaputaina ai. Ko te tuku tuwhera te tikanga o te kamupene mai i taua wā tonu.
Te wā o te MoE, me te huri ki ngā kaihautū
He nui ake te take o te aho here i tētahi putanga kotahi. Mai i te GLM-4.5, kua whakapaingia ia putanga mō te mea kotahi tonu: te noho mārama puta noa i ngā tekau karanga taputapu. He whāinga rerekē tēnā i te whakanui i ngā tatauranga paearu hurihanga-kotahi, ā, ka kitea i te iti iho o ngā tautohetohe taputapu kua hē te hanga, te iti iho o te kōrari i tētahi mahere roa, me te iti iho o te whanonga "ka tīmata anō whakapono i te mahi i te hurihanga 14" e nui ai te utu o ngā anga kaihautū.
02He aha a GLM-5.2, ā, i pēhea te ine i a ia
Ka whakaputaina e Z.ai ngā tatauranga paearu mō GLM-5.2 i te putunga GLM-5. E rua e whakaata ana i te mahi kaihautū tūturu, kaua i te mahara meka noa:
Terminal-Bench 2.1 — ngā mahi kaihautū rārangi-whakahau pae-roa. I pūrongotia e Z.ai.
SWE-bench Pro — te whakaoti take putunga-pū pono. I pūrongotia e Z.ai; i eke a GLM-5.1 ki te 58.4 i te whakamātautau ōrite.
He tau nā te whare rangahau ēnei, ehara i a mātou, ā, e whakaahua ana i te tūāpapa kāore i kōpakina. Ka whakatū rātou i te akomanga tauira e pā nei koe, kaua he whakaū.
He nui ake te uara o tētahi tau kotahi i tētahi paearu, nā te mea i ahu mai i tētahi kaiarotake kāore he take ki te whakapaipai i te tauira. I te Hōngongoi 2026, i arotakehia e te Center for AI Standards and Innovation a te National Institute of Standards and Technology o Amerika a GLM-5.2, ā, i kī rātou ko ia pea te tauira AI tuwhera-taumaha kaha rawa i wātea ai i tōna putanga, me ngā āheinga whānui e rite ana ki a GPT-5.2 me ngā āheinga rorohiko e rite ana ki a Claude Opus 4.6. I kitea hoki e taua arotake he iti iho ngā tono mataku ka araia e GLM-5.2 i ngā tauira tohutoro o Amerika, ā, ka taea te karo i ngā arai haumaru kua tō ki roto i ngā taumaha tuwhera ina whakahaerea aua taumaha e te tangata ake. Ka hoki te wāhanga 05 ki tēnā.
03He aha te tikanga o "3BIT-REAP", ā, he aha te utu ki a koe
REAP: te tapahi i ngā mātanga, kaua te whakakotahi
He 256 mātanga ia paparanga tā tētahi tauira MoE 744B-tawhā, e 8 e hohe ana mō tētahi tohu. Kāore te kaiwhakarite ara e tuari ōrite i te mahi — ka pāngia tonutia ētahi mātanga, ā, ka pāngia onoono tētahi hiku roa, he iti hoki tā rātou whai wāhi ina pēnā. Ka whakamahi te REAP (Router-weighted Expert Activation Pruning), nā Cerebras Research, i whakaaetia mō ICLR 2026, i tērā: ka whakatau ia mātanga mā tētahi paearu whai-take e whakakotahi ana i te kaha me te auau o te kōwhiritanga a te kaiwhakarite ara me te nui o tāna whai wāhi whakaputa, kātahi ka muku pūmau i ngā mea tatauranga-iti.
Ko te wāhanga whakamīharo ko te whakataurite ka toa e REAP. Ko te kōwhiringa mārama ki te muku i tētahi mātanga ko te whakakotahi ki tētahi kaiora, e rite ana ki te mea haumaru ake. Ko te kitenga a te pepa puta noa i ngā tauira MoE mai i te 20B ki te 1T tawhā, he kino ake te whakakotahi mō te mahi whakaputa, ā, ka mau tonu pai rawa te tapahi: tata-kore-ngaro i ngā paearu whakaputa waehere me te karanga taputapu ahakoa kua tangohia te haurua o ngā mātanga, i runga i ngā tūāpapa tae atu ki te Qwen3-Coder-480B me te Kimi-K2. Ko te ora tonu o te waehere me te karanga taputapu te mea tonu e hiahiatia ana kia pupurihia i tētahi tauira ko tōna take mō te ora he hangarau kaihautū.
Ngā taumaha 3-bit, me te utu whakawhiti pono
I tua atu i te tapahi, ka rongoātia ngā taumaha e ora tonu ana ki te 3 pae — he kaha ake i ngā taumata W4A16-AUTOROUND-REAP me W8A16 i ētahi atu wāhi o tā mātou rārangi, ā, kua rāhuitia mō ngā tūāpapa nunui rawa, i reira kāore he ara kē e taea ai te whakauru i tētahi tauira toru-hauwhā-taera-tawhā ki runga i tētahi pona whakahaere, me te pupuri tonu i te rere o ngā tohu i tētahi tere whaihua.
Ehara te kōpaki i te mea kore-utu. Ko te tauira hinengaro tika: ka whakahekea tuatahitia te hiku roa e te tapahi me te quantization. Ko ngā whanonga kiato, e mahia auau ana — te whai tohutohu, te hanganga waehere, te whakahōputu karanga taputapu, te whakamahere maha-hipanga — ka mau pai, koirā tonu te mea e whakaatu ana ngā hua REAP. Ko te mahara meka ngaro, ngā reo onoono me ngā meka rohe-iti ngā wāhi hei tūmanako i tētahi rerekētanga ki tētahi GLM-5.2 kāore i kōpakina. Mēnā ko tō kawenga mahi he "mōhio pū ki tētahi meka rerekē," utua he taumata pono-teitei ake. Mēnā he "whakahaere i tētahi mahere puta noa i ngā karanga taputapu e whā tekau, kāore e peka," koinei te taumata i hangaia ai te tikanga mō tēnā.
04Te take uara: he aha tonu te utu whakahaere o tētahi API LLM kore-ārai utu-iti
Ko te utu rārangi API i whakaputaina e Z.ai ake mō GLM-5.2 he $1.40 ia miriona tohu tāuru me te $4.40 ia miriona whakaputa. Ko tā mātou utu i runga i te tūāpapa ōrite he $0.73 / $2.34 — tata ki te 48% iti iho i te tāuru, 47% iti iho i te whakaputa. Koirā te utu-heke e hokona ai e te kōpaki, ā, kei te haere tahi me tētahi atu mea kāore e taea te hoko ā-kamupene tonu ahakoa te utu: kāore he paparanga whakakāhore.
I roto i tā mātou ake rārangi, ko te utu whakaranu te tau e whai take ana. Anei i te ōwehenga tāuru-ki-te-whakaputa 3:1, tata ki te āhua o tētahi kaihautū karanga-taputapu i te tikanga — he nui te horopaki me ngā hua taputapu e uru ana, he iti noa ngā tohu e puta mai ana.
| Tauira | Tāuru / 1M | Whakaputa / 1M | Whakaranu 3:1 | Tirohanga |
|---|---|---|---|---|
| GLM-5.2 · 3BIT-REAP | $0.73 | $2.34 | $1.13 | Kāo |
| Nemotron3Ultra · 3BIT-REAP | $0.75 | $3.30 | $1.39 | Kāo |
| Kimi-K2.6 · W4A16-AUTOROUND-REAP | $0.78 | $3.67 | $1.50 | Āe |
| DeepSeek-V4-Pro-0813 · 3BIT-REAP | $1.95 | $3.90 | $2.44 | Kāo |
| inkling · W4A16-AUTOROUND-REAP | $1.42 | $6.07 | $2.58 | Āe |
| Kimi-K3 · 3BIT-REAP | $3.83 | $19.12 | $7.65 | Āe |
E ono ngā tauira e noho ana i te papa $0.50 / $2.00 — MiniMax-M3, DeepSeek-V4-Flash, Laguna-S-2.1, MiMo-V2.5, MiMo-V2.5-Pro me Hy3 — he reiti whakaranu $0.875. He 29% i runga ake i taua papa a GLM-5.2, ā, ka hoko i a koe tētahi tūāpapa taumata-744B me tētahi haumitanga mau-roa ki te whanonga kaihautū e tautoko ana i a ia. Ko ngā mea katoa kua utua i runga ake i a ia, he 23% neke atu te nui o te utu whakaranu, ā, he 6.8× te nui o te utu o Kimi-K3. Koirā te tautohe uara i te rārangi kotahi: ko te wāhi utu-iti rawa o te rārangi hei hoko i tētahi tūāpapa kaihautū taumata-teitei.
Kua huri hei mahi
- Kotahi whakakī horopaki katoa — 262,144 tohu tāuru: $0.19. Ka pānui i tētahi putunga waehere nui katoa ki roto i te matapihi mō te iti iho i te rua tekau heneti.
- Kotahi whakaotinga roa-nui-rawa — 16,384 tohu whakaputa: $0.038.
- Kotahi hurihanga kaihautū — 20K uru, 2K puta: $0.019. He $0.115 te utu o taua hurihanga ōrite i runga i a Kimi-K3, he $0.047 i runga i a DeepSeek-V4-Pro.
- Kotahi rau hurihanga kaihautū: $1.93, hei whakataurite ki te $11.48 i runga i te tauira utu-nui rawa o te rārangi.
He taumaha-tāuru ngā kawenga mahi kaihautū, ā, kei reira tonu te painga nui rawa o GLM: he 5.2× iti iho tāna utu tāuru i tā Kimi-K3, he 2.7× iti iho i tā DeepSeek-V4-Pro. Mēnā ka tuku anō tāu kaihautū i tētahi kōrerorero e tipu haere ana ia hurihanga — ā, koinā te nuinga o ngā kaihautū — koinā te whakarea ka mutu i runga i tō nama.
05He aha te tikanga o "kore-ārai" i tēnei kūwaha
E rua ngā mea, tino tika: kāore he paparanga whakakāhore — kāore he kaiwehewehe e whakatau ana ko ēhea tono ka whakaaetia kia kitea e te tauira, kāore hoki he tauira kaupapa here e tuhi anō ana i tō tohutohu i mua i tōna taenga mai — me te kore o te tātari whakaputa, ko te tikanga ko ngā tohu i whakaputaina e te tauira ngā tohu ka riro i a koe, kāore i whakangaromia, kāore hoki i whakawhitiwhitia.
He kaupapa rerekē tēnā i tētahi API arumoni e panui ana i tētahi aratau "taritari" "kaiwhakawaehere" rānei. Ka whakahaerea tonu aua mea i te huarahi mā roto i te tū kaupapa here a te kaihoko; ko te mea ka panoni ko te paepae, kaua ko te noho o te kēti. I konei, kāore he kēti hei whakatika — ka whakahaerea te tauira i runga i tā mātou ake tautau GPU, ā, ko tōna whanonga te whanonga o te tūāpapa. Pānuitia te kitenga NIST CAISI o te wāhanga 02 hei kōrero hangarau, kaua hei whakatūpato, ā, ka whakamāramatia te hua: he tauira e noho ana tōna whakarōpū i roto i ōna taumaha, e whakahaerea ana me te kore o tētahi paparanga kaupapa here o waho, ka whanonga e ai ki ōna taumaha.
He aha i nui ai tēnei mō ngā kaihautū
He pātai ihirangi te tikanga o te whakakāhoretanga i te nuinga o te wā. I te mahi kaihautū, he pātai pono-whakahaere te nuinga. Ko tētahi whakakāhoretanga i te hurihanga 31 o tētahi rere 40-hurihanga ehara i te whakakāhoretanga ngāwari e pānuitia ana e koe ka haere tonu — he hipanga kua hē te hanga, he kaiwhakahaere kāore e mōhio me pēhea te haere tonu, he mahere pakaru, me ngā hurihanga 30 o ngā tohu kua pau me te kore mea hei whakaatu. Ka pāngia tonu tēnei e te rangahau haumarutanga, te tātari pūmate, te whakamātautau urutomo, te whakangungu whero-tīma, te pakimaero pakeke, te tuhi rongoā me te ture, me te hanga huinga raraunga mō te whakangungu haumarutanga, ā, he nui te utu o te hapa nā te mea ka tae mua i muri.
Ko te taha kē, kua noho nāu te paparanga kaupapa here. Ka whakahaerea te whakamahinga o te Kūwaha Shannon AI e tā mātou Kaupapa Here Whakamahi Haepapa; ko te tango i te whakatau a te kaihoko mai i te porowhita te tikanga me homai e koe tāu ake.
06He aha tonu ka homai e te kūwaha
| Āheinga | GLM-5.2 · 3BIT-REAP |
|---|---|
| Id tauira | GLM-5.2-3BIT-REAP |
| Matapihi horopaki | 262,144 tohu |
| Whakaputa nui rawa ia tono | 16,384 tohu |
| Streaming | Āe |
| Karanga taputapu | Āe |
Whakaaroaro / reasoning_effort | Āe |
JSON Schema (json_schema) | Āe |
response_format | Āe |
| Tirohanga / tāuru whakaahua | Kāo — kuputuhi anake |
| Paparanga whakakāhore / tātari whakaputa | Kāore |
Āta pānuitia te rārangi tirohanga i mua i tō hanga. He kuputuhi anake a GLM-5.2 i tā mātou kūwaha. Mēnā kei te hiahia koe i te tāuru whakaahua, e rima ngā kōwhiringa whai-tirohanga kei te rārangi — MiniMax-M3 me MiMo-V2.5 i te papa $0.50 / $2.00, Kimi-K2.6 i te $0.78 / $3.67, inkling i te $1.42 / $6.07, me Kimi-K3 i te $3.83 / $19.12. He tauira whai-take ko te tuku i te māramatanga whakaahua ki a MiniMax-M3, ā, ko ērā atu mea katoa ki a GLM-5.2, kāore he utu i tua atu i te huri i te aho tauira.
E toru ngā reo, kotahi te tauira
Ka taea te tae atu ki te tauira kotahi mā ngā hōputu tono e toru, ka streaming katoa: /v1/chat/completions (te reo OpenAI), /v1/messages (te reo Anthropic), me /v1/responses. Kāore tētahi mea e pā ana ki te tauira e panoni i waenga i a rātou — kōwhiria te mea e kōrero kē ana tō SDK, ā, ko te nuinga o te wā he panonitanga URL-tūāpapa kotahi noa koe e tata ana ki te whakahaere.
# OpenAI dialect — point your existing client at the gateway from openai import OpenAI client = OpenAI( base_url="https://api.shannon-ai.com/v1", api_key="sk-...", ) resp = client.chat.completions.create( model="GLM-5.2-3BIT-REAP", messages=[{"role": "user", "content": "..."}], tools=tools, # tool calling supported reasoning_effort="medium", # low | medium | high stream=True, )
Kei te tuhinga API a Shannon AI ngā hanganga tono me te whakautu mō ngā reo e toru, te hōputu karanga-taputapu, ngā āhua kaupapa streaming, me tētahi papa tākaro ā-taunekeneke.
07Ā tēhea wā ka kōwhiri i a GLM-5.2 — ā tēhea wā kāore
Toro atu ina
- Kei te whakahaere koe i ngā kaihautū. Ko ngā porowhita karanga-taputapu roa te mea i hangaia ai me te whakangungu-whai-muri tonutia ai tēnei whakapapa, ā, mā te utu tāuru ka taea te tuku anō i tētahi kōrerorero e tipu haere ana ia hurihanga.
- Kei te mahi koe puta noa i tētahi putunga waehere nui. He horopaki 262K, he hōputu karanga-taputapu pono, ā he $0.19 ki te pānui i te matapihi katoa.
- Ko te ranea tāu rohe. Te whakarōpū, te tango, te hanga raraunga waihanga, te tuhi anō pukuputu — ngā mahi katoa e whakatauria ai te oranga o tētahi kaupapa mā te utu ia-karanga whakarea ki tētahi tau nui.
- Kei te pakaru tō ara mā ngā whakakāhoretanga. Te rangahau haumarutanga, te whakamātautau whakahoariri, te pakimaero kore-ārai, te tuhinga mataku — kāore te momo hapa nei kei konei.
Toro atu ki tētahi atu ina
- Kei te hiahia koe i te tirohanga. He kuputuhi anake. Tirohia ngā kōwhiringa whai-tirohanga kei runga ake nei.
- Kei te hiahia koe i te whakaaroaro hurihanga-kotahi hōhonu rawa. Mō ngā raru wetewete pakeke rawa kotahi-hurihanga, ko DeepSeek-V4-Pro-0813 · 3BIT-REAP te taputapu taumaha ake i te 2.2× te utu whakaranu — mō ngā karanga e noho ana te tuanui hei rohe, koirā te whakawhiti tika.
- Kei te hiahia koe i te pono nui rawa mō te mātauranga ngaro. He iti rawa ake te kōpaki o te taumata
W8A16— MiMo-V2.5-Pro me Hy3 — ā, e noho ana i te papa $0.50 / $2.00, i runga i ngā tūāpapa iti ake.
Ko ia tatauranga tūāpapa o tēnei whārangi i whakaputaina e tētahi atu, ehara i a mātou. Ko ngā tatauranga utu me te āheinga he uara nā tā mātou ake rārangi ora, ka taea te whakaū ki te /v1/models mā tō kī ake. Tirotirohia ngā puna tuatahi e koe ake.
He tatauranga i pūrongotia e Z.ai mō te tūāpapa kāore i kōpakina ngā tatauranga paearu GLM-5.2 (Terminal-Bench 2.1, SWE-bench Pro); kei te tatari tonu te tāruatanga motuhake. Ka whakaaturia ngā tatauranga Claude Opus 4.8 me GPT-5.5 hei tohutoro anake. Ko te utu rārangi API GLM-5.2 ā-kamupene tonu e whakaputaina ana e Z.ai, Hepetema 2026.
08Ngā pātai auau
He aha te API GLM kore-ārai i runga i a Shannon AI?
Ko te id tauira GLM-5.2-3BIT-REAP i runga i te Kūwaha Shannon AI: he hanga kua kōpakina o te tūāpapa kaihautū GLM-5.2 a Z.ai, e whakahaerea ana i runga i tā mātou ake tautau GPU, kāore he paparanga whakakāhore, kāore hoki he tātari whakaputa. He $0.73 te utu ia miriona tohu tāuru me te $2.34 ia miriona tohu whakaputa, he matapihi horopaki 262,144 tohu tāna, ā, ka tae atu i te /v1/chat/completions, /v1/messages, me te /v1/responses.
He aha ai he utu-iti ake a GLM 5.2 kore-ārai i ērā atu tauira nunui o te rārangi?
Nā te āhua o tōna kōpaki. Ka whakakotahi te 3BIT-REAP i ngā taumaha 3-bit me te REAP, he tikanga tapahi-mātanga mō ngā tauira Whakaranu-o-Ngā-Mātanga e muku ana i ngā mātanga onoono-arataki i mua i te quantizing. He tūāpapa MoE korekore me te tata 40B tawhā hohe ia tohu, kua tapahia, kua quantized hoki i tēnei ara, he iti rawa ake tōna whakamahi mahara, he iti rawa ake hoki ōna hēkona-GPU ia tohu i tētahi tauira kiato o te kounga ōrite, ā, ka whakaatahia tērā e te utu.
Ko te tikanga o te GLM kāore he whakakāhore, kāore rawa he arai haumaru i te tauira?
Ko te tikanga kāore a Shannon e tāpiri i tētahi. Kāore he kaiwehewehe whakakāhore kei mua i te tauira, kāore hoki he tātari i ngā tohu e puta mai ana i a ia. Ka pupuri tonu te tūāpapa i ōna āhuatanga i homai e tāna ake whakangungu, engari kāore he mea o waho e whakakore ana i ērā. Nāu te paparanga kaupapa here, ā, kei raro te whakamahinga i te Kaupapa Here Whakamahi Haepapa a Shannon AI.
Ka tautoko a GLM-5.2 3BIT-REAP i te tirohanga, i te tāuru whakaahua rānei?
Kāo. He kuputuhi anake tēnei tauira. Mēnā kei te hiahia koe i te tāuru whakaahua i runga i tētahi kūwaha kore-ārai, whakamahia tētahi o ngā tauira whai-tirohanga o te rārangi pēnei i a MiniMax-M3, Kimi-K2.6, Kimi-K3, MiMo-V2.5, inkling rānei. Ka tautoko a GLM-5.2 i te putanga hanganga — te json_schema me te response_format — me te karanga taputapu.
He aha te matapihi horopaki me te roa whakaputa nui rawa?
262,144 tohu horopaki, 16,384 tohu whakaputa mō ia tono. Ko te whakakī i te matapihi horopaki katoa kotahi tau he tata 19 heneti i te $0.73 ia miriona tohu tāuru, ā he iti iho i te 4 heneti te utu o tētahi whakaoti roa-nui-rawa.
He tauira kaihautū kore-ārai pai tēnei mō ngā porowhita karanga-taputapu roa?
Koinā te mea ka toro tuatahi atu mātou. Kua whakangungua mārōrō te whakapapa GLM mō te mahi kaihautū mai i te GLM-4.5, e tautokona ana te karanga taputapu me te reasoning_effort, ā, ko te tikanga o te utu he tata rua tāra mō te rau hurihanga kaihautū 20,000-tohu, kaua te tekau mā tahi tāra. He rahunga pono hoki ngā whakakāhoretanga i roto i ngā porowhita kaihautū, ā, kāore he paparanga whakakāhore i konei hei whakaoho i tētahi.
Whakamātauhia te API GLM kore-ārai
$0.73 tāuru · $2.34 whakaputa · horopaki 262K · kāore he paparanga whakakāhore.
Pānuitia ngā Tuhinga API Tirohia Ngā Tauira KatoaShannon Lab LLC, New Mexico, USA · e whakahaerea ana te whakamahinga e te Kaupapa Here Whakamahi Haepapa
Puna kōrero: Z.ai GLM-5 repository · Du et al., GLM: General Language Model Pretraining with Autoregressive Blank Infilling, ACL 2022 · Zeng et al., GLM-130B: An Open Bilingual Pre-trained Model · Lasby et al., REAP the Experts: Why Pruning Prevails for One-Shot MoE Compression, ICLR 2026 · NIST CAISI Assessment of Z.ai's GLM-5.2, July 2026. He tatauranga paearu tūāpapa i pūrongotia e te whare rangahau, ā he mea whakaharatau tonu; e whakaahua ana i te GLM-5.2 kāore i kōpakina, kaua i tēnei hanga kua kōpakina. He uara rārangi a Shannon AI ngā tatauranga utu, horopaki me te āheinga, e rite ana ki te 5 o Hepetema 2026.