Hacker News

Qwen3.5: Ɛkɔ Native Multimodal Agents so

Qwen3.5: Ɛkɔ Native Multimodal Agents so Saa nhwehwɛmu yi kɔ akyiri wɔ qwen3 mu, hwehwɛ nea ɛkyerɛ ne nkɛntɛnso a ebetumi aba mu. Nsusuwii Titiriw a Wɔakata So Saa nsɛm yi hwehwɛ: Nnyinasosɛm ne nsusuwii atitiriw Nea mfaso wɔ so...

12 min read Via qwen.ai

Mewayz Team

Editorial Team

Hacker News
Afei ma menyɛ blog post no. Megyina nsɛm a ɔmanfoɔ wɔ a ɛfa Qwen3.5 ho a ɛfiri Alibaba Qwen kuo (a wɔyii no adi wɔ afe 2025) mu so, mɛkyerɛw SEO asɛm a ɛyɛ pɛpɛɛpɛ, a ɛkɔ akyiri. HTML nipadua mu nsɛm a edi mũ ma blog post no ni: --- .

Qwen3.5: Ɛkɔ Native Multimodal Agents so

Qwen3.5 gyina hɔ ma Alibaba Cloud no ahurututu a ɛyɛ aniberesɛm sen biara wɔ AI mu de besi nnɛ — abusua bi a wɔde fapem nhwɛso ahorow a wɔasisi afi fam de adi nsɛm, mfonini, ɔdio, ne video ho dwuma wɔ nhyehyɛe biako a wɔaka abom mu. Sɛ anka wɔbɛbɔ multimodal tumi ahorow agu kasa nkutoo akyi dompe so no, Qwen3.5 di modality biara ho dwuma sɛ ɔman ba a odi kan, na ɛma AI adwumayɛfo kuw foforo a wotumi hu, te, kenkan, na wɔyɛ ade wɔ kurom hɔ no tumi yɛ adwuma.

Dɛn na Ɛma Qwen3.5 yɛ "Native" Multimodal Model?

Awo ntoatoaso a atwam a ɛwɔ multimodal AI no taa de wɔn ho to adapter layers so — encoder ahorow a ɛtetew mu ma anisoadehu anaa ɔdio a wɔapam wɔ kasa nhwɛso kɛse bi so wɔ ntetee akyi. Qwen3.5 gyae saa nhwɛsoɔ no. Ne nhyehyeɛ no yɛ natively multimodal, a ɛkyerɛ sɛ model no bom sua representations wɔ text, image, audio, ne video nyinaa mu wɔ pre-training mu sen sɛ ɛnam post-hoc alignment so.

Saa nhyehyeɛ a wɔpaw yi wɔ nkyerɛkyerɛmu titire. Esiane sɛ modalities nyinaa kyɛ transformer backbone ne attention mechanism koro nti, model no nya cross-modal ntease a ɛyɛ den. Ebetumi asusuw nhyehyɛe bi a ɛwɔ PDF mu ho bere a bere koro no ara ɛrekyerɛw akwankyerɛ a wɔaka afa saa nhyehyɛe no ho — a nsɛm a ɛyɛ den a nhyehyɛe ahorow a egyina adapter so de ba no nni mu. Nea efi mu ba ne outputs a ɛyɛ mmerɛw, ɛne ne ho hyia bere a nnwuma fa input ahorow pii ho prɛko pɛ.

Alibaba Qwen kuw no ayi Qwen3.5 adi wɔ parameter akɛseɛ pii mu, a ɛtoa open-weight atetesɛm a ɛmaa Qwen a wɔayi no adi kan no agye din wɔ developers ne nnwuma nyinaa mu. Saa kwan a wɔfa so kɔ hɔ yi ho hia: ɛma nnwuma akɛseɛ nyinaa tumi siesie na wɔde multimodal agents a tumi wom di dwuma wɔ wɔn ankasa infrastructure so.

Ɔkwan Bɛn so na Qwen3.5 Ma AI Agent Tumi Nya Nkɔso?

Asɛmti ketewa "Towards Native Multimodal Agents" kyerɛ nsakrae a wɔahyɛ da ayɛ wɔ sɛnea yesusuw mfonini akɛse ho no mu. Qwen3.5 nyɛ chatbot a ɛtumi hwɛ mfonini kɛkɛ — ɛyɛ agent framework. Nhwɛsoɔ no de adwinnadeɛ a wɔde di dwuma ho nsusuiɛ a wɔasisi mu, dwumadie frɛ, ne nhyehyɛeɛ a wɔde yɛ adwuma a ɛma ɛyɛ adwuma wɔ ne ho wɔ adwumayɛ nhyehyɛeɛ a ɛyɛ den mu.

Tumi titiriw a ɛkyerɛkyerɛ Qwen3.5 agent suban mu no bi ne:

  • Multi-turn tool orchestration: Qwen3.5 tumi yɛ nhyehyɛe na ɛyɛ anammɔn pii nnwuma denam nkɔnsɔnkɔnsɔn API frɛ, database nsɛmmisa, ne koodu dwumadie so — asiesie ne nhyehyeɛ wɔ berɛ ankasa mu a egyina ntamgyinafoɔ aba so.
  • Aniwa so grounding ne GUI nkitahodi: Model no tumi kyerɛ screenshots ase, kyerɛ UI elements, na ɛma click anaa input nneyɛe a ɛyɛ pɛpɛɛpɛ, bue ɔpon ma browser-based ne desktop automation agents.
  • Nsusuwii a ɛfa nsɛm a ɛfa ho tenten: Ɛnam nsɛm a ɛfa ho mfɛnsere a wɔatrɛw mu nti, Qwen3.5 di nkrataa atenten, video a wɔatrɛw mu nnidiso nnidiso, ne nkɔmmɔbɔ a ɛkyɛ ho dwuma a ɛnhwere nkitahodi anaasɛ ne werɛ mfi akwankyerɛ a atwam.
  • Nsusuwii akwan a wɔde afrafra: Sɛ yɛde adwene-kwan foforo a efi Qwen3 so si so a, nhwɛsode no betumi asesa mmuae a ɛkɔ ntɛmntɛm, a ɛyɛ mmerɛw ne nsusuwii a emu dɔ, nkɔnsɔnkɔnsɔn mu nsusuwii a egyina adwuma a ɛyɛ den so.
  • Kasa pii ne mmara a wɔka no yiye: Adwumayɛ a emu yɛ den wɔ kasa ne nhyehyɛe nhyehyɛe du du pii mu ma Qwen3.5 yɛ nea mfaso wɔ so ma wiase nyinaa adwumayɛbea a wɔde di dwuma ne developer nnwinnade.

Saa tumi yi bom ma Qwen3.5 fata ma wiase ankasa mu agent deployments — efi automated customer support systems a ɛkenkan nkrataa na ɛhwɛ screen recordings, kosi nhwehwɛmu aboafo a wɔboaboa nsɛm ano wɔ text, charts, ne audio interviews so.

Dɛn Nti na Native Multimodality Ho Hia Ma Adwumayɛ Dwumadie?

Wɔ nnɛyi nnwuma fam no, data ntaa mma wɔ ɔkwan biako so. Adetɔn kwan bi fa imel (nsɛm), nneɛma a wɔyɛ ho ɔyɛkyerɛ (video), apam a wɔde wɔn nsa ahyɛ ase (mfonini a wɔa scan), ne wɔn a wɔde wɔn ho hyɛ mu frɛ (adio) ho. Amanneɛ kwan so AI nnwinnade hyɛ akuw ma wɔde nhwɛso ahorow a ɛsono emu biara di dwuma ma modality biara, na ɛyɛ adwumayɛ nhyehyɛe a emu apaapae ne nkabom so.

a wɔde ahyɛ mu

Native multimodal models te sɛ Qwen3.5 yi hia a ɛho hia sɛ wɔpam AI nnwinnade a wɔde di dwuma biako bom no fi hɔ. Sɛ model biako tumi kenkan wo invoices, hwɛ wo ntetee video ahorow, na ɔkyerɛw wo nhyiam ahorow no a, automation stack no nyinaa hwe ase kɔ layer biako a wotumi de ho to so kɛse — na ɛhɔ na adwumayɛ mu mmɔdenbɔ ankasa fi ase.

💡 DID YOU KNOW?

Mewayz replaces 8+ business tools in one platform

CRM · Invoicing · HR · Projects · Booking · eCommerce · POS · Analytics. Free forever plan available.

Start Free →
na ɛkyerɛ sɛ woayɛ

Saa nkabom yi ho hia wɔ nsenia mu. Nnwuma a wɔreyɛ adwuma wɔ platforms te sɛ Mewayz — a ɛka adwumayɛ module 207 bom dedaw fi CRM so kosi project management so — te tumi a ɛwɔ biribiara a ɛwɔ beae biako no ase. Sɛ AI di nyansapɛ koro no ara akyi a, mfaso a ɛwɔ compounding efficiency mu no yɛ kɛse. Sɛ anka wɔbɛhwɛ AI adetɔnfoɔ baanum so no, akuo bɛtumi de multimodal backbone baako a ɛdi nkrataa dwumadie, aniwa so nhwehwɛmu, adwuma a egyina nne so, ne amanneɛbɔ a nyansa wom ho dwuma wɔ pipeline baako mu.

Sɛ dɛn na Qwen3.5 Toto Frontier Models Afoforo ho?

AI space a ɛwɔ multimodal wɔ afe 2025 ne afe 2026 mu no abɛyɛ akansi kɛseɛ. OpenAI GPT-4o, Google Gemini 2.0 abusua, ne Anthropic Claude mfonini ahorow nyinaa de akwan horow pii tumi ma. Baabi a Qwen3.5 da nsow ne ne mu duru a wɔabue, native (ɛnyɛ bolted-on) multimodality, ne strong agentic tool-use out of the box.

Benchmark aba kyerɛ sɛ Qwen3.5 resi akan wɔ soro anaa ɛbɛn soro wɔ gyinapɛn nhwehwɛmu ahorow nyinaa mu wɔ kasa ntease, akontaabu mu nsusuwii, mmara awo ntoatoaso, mfonini ntease, ne video ntease mu. Ebia nea ɛho hia kɛse ma nnwuma a wɔfa no, tumi krataa a wɔde ma a wɔabue mu no kyerɛ sɛ ahyehyɛde ahorow betumi ayɛ Qwen3.5 wɔ ankorankoro nnwuma so — mfaso a ɛyɛ gyinaesi ma nnwuma a wɔwɔ data tumidi ahwehwɛde a emu yɛ den te sɛ sikasɛm, akwahosan, ne aban.

Model no agentic design nyansapɛ nso ma ɛda nsow. Bere a akansifo pii di mu wɔ asɛmmisa mmuae a wɔde dannan biako mu no, wɔayɛ Qwen3.5 ama adwuma a ɛkɔ so daa, a ɛdannan pii wɔ baabi a nhwɛso no kura tebea mu, de nnwinnade di dwuma, na ɛsakra ne nhyehyɛe wɔ nkitahodi a wɔatrɛw mu.

Dɛn na Daakye Bɛfa Multimodal AI Agents?

Qwen3.5 nyɛ awiei na mmom ɛyɛ trajectory marker. "Towards" a ɛwɔ ne subtitle mu no yɛ nea wɔahyɛ da ayɛ — yɛda so ara wɔ ti ahorow a edi kan no mu wɔ nea native multimodal agents bɛyɛ ho. Ɛda adi sɛ nkɔso a ɛbɛba bere tiaa bi no bi ne robɔt ne honam fam-wiase sensor ahorow a wɔde bɛka abom kɛse, bere ankasa mu nsu a ɛsen fa akwan horow so nkitahodi, ne nkae ne nhyehyɛe nhyehyɛe a ɛyɛ nwonwa kɛse a ɛma adwumayɛfo di nnwuma a ɛkyɛ adapɛn pii ho dwuma wɔ wɔn ankasa mu.

Wɔ nnwuma fam no, ade a mfaso wɔ so a wode bɛkɔ no da adi pefee: ɛsɛ sɛ nnwinnade a wopaw nnɛ no yɛ krado ma AI-native adwumayɛ ɔkyena. Platform ahorow a ɛde adwumayɛ adwumayɛ nhyehyɛe ahorow no si mfinimfini dedaw no de wɔn a wɔde di dwuma no si hɔ ma wɔde multimodal agents hyɛ mu a ɛnyɛ den, sen sɛ wɔbɛsan de nhyehyɛe ahorow a wɔatwa mu no ahyɛ mu wɔ nokwasɛm no akyi.

Nsɛmmisa a Wɔtaa Bisa

So Qwen3.5 yɛ nea wɔabue ano na wɔde di dwuma kwa?

Qwen3.5 yɛ nea Alibaba Cloud Qwen kuw no ayi no adi sɛ open-weight model, a ɛtoa ɔkwan a wɔde Qwen2 ne Qwen3 sii hɔ no so. Model weights no wɔ hɔ a wobetumi atwe kwa na wobetumi de adi dwuma wɔ private infrastructure so. Tumi krataa ho nsɛm pɔtee no gu ahorow wɔ nhwɛsode kɛse mu, enti ɛsɛ sɛ nnwumakuw hwɛ tumi krataa no mu ma nea wɔapaw no, nanso Qwen ntoatoaso no aka ɔhye so nhwɛso mmusua a wɔama ho kwan kɛse no ho, a ɛboa nhwehwɛmu ne aguadi mu dwumadie nyinaa.

Ɔkwan bɛn so na ɛsono Qwen3.5 wɔ Qwen3 ho?

Bere a Qwen3 de afrafra adwene akwan ne kasa-ne-nsusuwii tumi a emu yɛ den bae no, Qwen3.5 ma adansi no so kɔ native multimodality so. Wei kyerɛ sɛ wɔfa nhwɛsoɔ a wɔaka abom so di nsɛm, mfonini, ɔdio, ne video ho dwuma firi nteteeɛ a ɛdi kan — wɔmfa nka ho sɛ tumi a ɛtɔ so mmienu. Qwen3.5 nso hyɛ agentic features te sɛ adwinnadeɛ a wɔde di dwuma, dwumadie frɛ, GUI nkitahodiɛ, ne anammɔn pii adwuma nhyehyɛeɛ mu den kɛseɛ, na ɛma ɛyɛ atirimpɔw-a-wɔasi ama autonomous AI agent adwumayɛ.

So metumi de Qwen3.5 ahyɛ me adwumayɛ kwan a ɛwɔ hɔ dedaw no mu?

Yiw. Qwen3.5 boa standard API-based deployment na ɛne som frameworks a agye din te sɛ vLLM, Ollama, ne Hugging Face Transformers hyia. Wɔ nnwuma a wɔde dwumadie nhyehyɛeɛ a ɛwɔ ne nyinaa mu te sɛ Mewayz redi dwuma dedaw no, wɔbɛtumi de multimodal AI tumi ahyɛ module a ɛwɔ hɔ dada mu — a ɛbɛma nkrataa nhwehwɛmu ayɛ adwuma wɔ wo CRM mu, ayɛ nhumu afiri nsɛm a wɔde akɔ so wɔ adwuma no sohwɛ mu, anaasɛ ɛbɛma adetɔfoɔ nkitahodiɛ a nyansa wom anya ahoɔden wɔ akwan ahodoɔ so.


Nsakraeɛ a ɛrekɔ native multimodal AI agents so no reyɛ ntɛmntɛm, na nnwuma a ɛwɔ gyinabea pa a ɛbɛma wɔanya so mfasoɔ ne wɔn a wɔreyɛ adwuma dedaw afiri platform a wɔaka abom so. Mewayz de module 207 — efi CRM ne invoicing so kosi project management ne marketing automation — ba adwumayɛ OS biako a nnipa bɛboro 138,000 gye di mu. Si wo AI-asiesie adwumayɛ nnɛ. Fi ase ne Mewayz na hwɛ sɛnea adwumayɛ nhyehyɛe a wɔaboaboa ano ma AI awo ntoatoaso a edi hɔ no a wobegye atom no yɛ nea ɛnyɛ den.

Try Mewayz Free

All-in-one platform for CRM, invoicing, projects, HR & more. No credit card required.

Start managing your business smarter today

Join 30,000+ businesses. Free forever plan · No credit card required.

Ready to put this into practice?

Join 30,000+ businesses using Mewayz. Free forever plan — no credit card required.

Start Free Trial →

Ready to take action?

Start your free Mewayz trial today

All-in-one business platform. No credit card required.

Start Free →

14-day free trial · No credit card · Cancel anytime