Google компани үр ашиг, чанарын тэнцвэрийг хангасан Gemini 3.6 Flash загвараа зах зээлд нэвтрүүлж, өрсөлдөөнд эргэн ирлээ.
Мягмар гарагт танилцуулсан Gemini 3.6 Flash нь өмнөх 3.5 Flash загвартай харьцуулахад гаралтын токен зарцуулалтыг 65 хуртэлх хувиар бууруулж, нийт дүнгээрээ 17 хувийн хэмнэлт үзүүлжээ. Энэхүү өөрчлөлт нь хиймэл оюун ухааны ашиглалтын өртөг нэмэгдэж буй өнөө үед чухал ач холбогдолтой юм. Гэсэн хэдий ч гүйцэтгэлийн хувьд Anthropic-ийн Claude Sonnet 5, OpenAI-ийн GPT-5.6 болон Grok 4.5 загваруудаас хоцорч байгаа нь уг загварыг зах зээлд өөрийн байр сууриа олоход хүндрэлтэй болгож байна.
Тус компани мөн Gemini 3.5 Flash-Lite болон Flash Cyber гэсэн хоёр шинэ хувилбарыг танилцууллаа. Flash-Lite нь хамгийн хурдан бөгөөд зардал багатай загвар болж буй бол Flash Cyber нь кибер аюулгүй байдлыг хангах зорилгоор засгийн газар болон итгэмжлэгдсэн түншүүдэд зориулан бүтээгджээ. Энэхүү загвар нь “CodeMender” аюулгүй байдлын агентаар дамжуулан эмзэг байдлыг илрүүлэх, засах туршилтын хөтөлбөрт ашиглагдах юм.
Одоогийн байдлаар Gemini 3.6 Flash болон 3.5 Flash-Lite загваруудыг байгууллагын хэрэглэгчид болон Gemini аппликейшнээр дамжуулан ашиглах боломжтой байна. Flash-Lite загвар нь удахгүй Google хайлтын системд нэмэгдэхээр төлөвлөгджээ. Хэдийгээр Google өөрийн загваруудаа шинэчилж байгаа ч илүү хүчирхэг “Pro” хувилбар болон дараа үеийн Gemini 4 загварыг хөгжүүлж байгаа нь энэхүү шинэчлэлтийг томоохон дэвшил гэхээсээ илүүтэйгээр ирээдүйн бүтээгдэхүүнүүдээ танилцуулах хүртэлх түр зуурын алхам гэж харагдаж байна.
Дэлгэрэнгүйг эх сурвалжаас харах
Эх сурвалжийг нээх ↓
Remember when Google released its Gemini 3 AI model last year and it seemed like it was going to force itself into the conversation among the frontier labs with major benchmarking achievements and major fanfare around its image generation model? That all faded mighty fast, huh? Well, Google is back to remind everyone that it is competing—though it’s not exactly blowing anyone away.
On Tuesday, the company introduced Gemini 3.6 Flash, the “workhorse” version of the company’s flagship model that is supposed to provide the best balance of quality and efficiency. And it does seem to eat up fewer tokens, which is currently a meaningful benchmark in the post-tokenamxxing era. Per Google, Gemini 3.6 Flash cuts output token by as much as 65% compared to 3.5 Flash in some uses, and is uses 17% fewer output tokens overall than the previous model.
That’s noteworthy as people become more aware of the costs associated with AI use, but the performance of Gemini 3.6 Flash isn’t going to blow anyone away. It appears to be behind Anthropic’s Claude Sonnet 5 and OpenAI’s GPT-5.6 in most major benchmarking tests, and it’s even started to lag behind the latest Grok model, 4.5, in tasks like agentic coding (Grok can probably thank the Cursor team for the boost there, but the score is the score). Considering that Gemini 3.6 Flash isn’t that much cheaper than most of its competitors—it’s about in line with Grok 4.5 and GPT-5.6—it’s going to have a hard task carving out a niche in the model war.
More Gemini 3.5 models released as well
In addition to Gemini 3.6 Flash, Google also introduced two other models: 3.5 Flash-Lite and 3.5 Flash Cyber—both bolstering the previous iteration of the company’s flagship model. Google says 3.5 Flash-Lite is its “fastest, most cost-effective” model. Like 3.6 Flash, it builds on the performance of its past version while improving overall efficiency.
3.5 Flash Cyber, meanwhile, is Google’s foray into the cybersecurity-specific model. While the company isn’t shrouding the model in the doomsday language that Anthropic used to introduce Mythos—a model so dangerous in the company’s estimation that it can only be made available to a limited number of partners in a very restricted capacity—it is keeping Flash Cyber on rails. Per the company, it’s making the model “exclusively available to governments and trusted partners” via its CodeMender AI security agent as part of a pilot program meant to detect and patch security vulnerabilities.
While Google has moved the ball forward on its model development with 3.6 Flash, the company is still building up Gemini 3.5, and is promising a Pro version, its most powerful model, soon. The company is also pre-training Gemini 4, presumably, so while Gemini 3.6 might be an improvement, it seems it’s really more just an advertisement to remind you Gemini exists until the real updates come.
Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are both available starting today for Gemini enterprise users and people using the Gemini app. 3.5 Flash-Lite will also be rolling out to Google Search soon.
