Bollywood HungamaEXCLUSIVE: Gulshan Devaiah DEBUNKS Wikipedia's claim that he made his debut in Salman Khan's Dil Ne Jise Apna Kahaa: “Even I have seen it and I laugh at it!”Daily MaverickWHAT’S COOKING: Bockwurst with a warm potato and bacon salad and braised red cabbagePunchThree defraud Delta community leader of N5.4mThe Jerusalem PostTunisia increases prison sentences for defendants in deadly 2023 Djerba synagogue attack - reportESPNExpert picks for the East, West and NBA championsCNN TürkSON DAKİKA... Afyonkarahisar'da kafa kafaya facia: 4 can kaybıSouth China Morning PostFor Filipinos, virtual assistant trumps doctor as ‘dream job’. But what’s the reality?ColliderChris Pratt’s Official ‘Reacher’ Replacement Surges on Prime Video 4 Years LaterZDF heuteEntdecken Sie das ZDF-NachrichtenstudioThe Hollywood ReporterHiam Abbass to Receive European Achievement in World Cinema AwardBillboardSam Fender Is Doing a Proper Massive Stadium Tour for 2027 in U.K. & Ireland, Including Nights at WembleyVilaWebIlla diu que l’amnistia tanca una etapa i Junts i ERC li recorden que s’hi va oposar
The Daily Newsstand · Free, Always
Wednesday, October 21, 2026

Qwen3.8-Flash-Nextと競合するオープンモデル「Ornith-1.5-122B」が登場

Translate

AI研究組織のOrnithがAIモデル「Ornith-1.5-122B」を2026年10月20日に公開しました。同等規模のオープンモデルと比べて高性能であることがアピールされています。

Aloha!🌺 Meet Ornith-1.5-122B: No Tradeoff Between Strong and Fast

Stronger. Faster. Leaner. Ornith-1.5-122B delivers frontier-level capabilities with exceptional inference efficiency. With superior instruction following, stronger task decomposition, and deeper understanding… pic.twitter.com/vyyglS8Dsp

— Ornith (@ornith_) October 20, 2026

Ornith-1.5-122Bは総パラメーター数1220億のMoEモデルで、80GBのVRAMを搭載した単一のNVIDIA DGXシステムで実行可能です。

「Ornith-1.5-122B」「Qwen3.8-Flash-Next(125B)」「Qwen3.8 27B」「Nemotron 3.5 Super VL(120B)」「DeepSeek-V4-Flash-0731(284B)」「Beam(501B)」の各種ベンチマーク結果をまとめた図が以下。Ornith-1.5-122Bは同等規模のQwen3.8-Flash-NextやNemotron 3.5 Super VLをほとんどのテストで上回りました。一部のテストではDeepSeek-V4-Flash-0731より高いスコアを記録しています。

View the original on GIGAZINE →

KioskNews shows a cleaned-up reading view extracted from the publisher’s page — the original always lives on their site, not ours.