On the 21st (local time), Google unveiled three new lightweight versions of its artificial intelligence (AI) model "Gemini." It again did not release the top-tier model, "Gemini 3.5 Pro." As Google's next-generation model rollout is delayed, concerns are growing that it may be falling behind OpenAI and Anthropic in the AI race.
Google DeepMind said it released three new lightweight AI models: "Gemini 3.6 Flash," "3.5 Flash-Light," and "3.5 Flash Cyber," which enhances cybersecurity capabilities.
Gemini 3.6 Flash is a follow-up to Gemini 3.5 Flash introduced in May, featuring improved multimodal performance in coding, office and knowledge work, and voice and images. It also increased efficiency by reducing output token (the basic unit an AI model uses to process information and generate responses) usage by 17%.
Gemini 3.5 Flash-Light is a model specialized for AI agent tasks. With up to 350 output tokens per second, it is optimized for AI agent work that requires minimizing computational latency.
The cybersecurity-specialized model "3.5 Flash Cyber" can quickly find and fix security vulnerabilities, so it is offered only to governments and a limited number of trusted corporations and institutions.
However, Google again did not release the much-anticipated "Gemini 3.5 Pro." Gemini 3.5 Pro is Google's top-tier model, strong in complex reasoning, coding, and advanced math. Chief Executive Officer Sundar Pichai had earlier said the planned release window for Gemini 3.5 Pro was June, but as of mid-July it still has not been unveiled.
Bloomberg News previously reported that Google has been struggling to improve Gemini's coding capabilities.
On the day, Google said, "Gemini 3.5 Pro is currently being tested with partners and will be widely released once it is ready," adding, "We are already focusing on developing the next-generation model Gemini 4."
In the industry, there are growing concerns that Google may again be falling behind in the AI performance race. Recently, Anthropic introduced "Mythos 5" and "Fable 5," and OpenAI released "GPT-5.6," intensifying the battle for the lead. SpaceXAI and Meta, which had been seen as trailing Google, also released "Grok 4.5" and "Muse Spark 1.1" recently, and in terms of performance they have surpassed Google's Gemini.
In the AI sector, new models are pouring out every one to two months, so delays in launching cutting-edge models can lead to weakened technological competitiveness.
Chinese corporations have joined as well. Moonshot AI's "Kimi K3" shocked the industry with performance comparable to Anthropic and OpenAI models, and Zhipu AI's "GLM-5.2" also scored higher than Gemini in performance evaluations.
AI evaluation firm Artificial Analysis rated Gemini 3.6 Flash's intelligence metric at 50 points, the same as its predecessor, placing it 12th—below GLM-5.2 and Muse Spark 1.1 (51 points each).