Google product management senior director Tulsee Doshi published an official blog post on July 21, announcing the launch of Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber—each designed specifically for AI agent workflows. In addition, Gemini 4 pre-training work has already started.
Gemini 3.6 Flash Specs: OSWorld 83.0%, 17% fewer output tokens
Gemini 3.6 Flash is the flagship work model in the new series, with notably improved performance in code writing, knowledge work, and multimodal tasks. Computer use capability evaluation (OSWorld-Verified) reached 83.0%. According to the Artificial Analysis Index, its output token usage is 17% lower than its predecessor, with up to a 65% reduction in benchmark tests such as DeepSWE.
Computer use tools are now directly built into the Gemini API and the enterprise edition, with no additional configuration required. Pricing: $1.50 per 1 million tokens for input, and $7.50 per 1 million tokens for output.
Gemini 3.5 Flash-Lite Specs: 350 tokens/s output speed and adjustable thinking levels
Gemini 3.5 Flash-Lite is designed for low-latency and high-throughput tasks (such as agent search and document processing). Key specs are as follows:
Output speed: 350 tokens/s
Pricing: $0.30 per 1 million tokens for input, $2.50 per 1 million tokens for output
Thinking levels: supports adjustable thinking levels, allowing developers to switch between a low-thinking mode (suitable for high-frequency tasks) and a high-thinking mode (handling complex sub-agent workflows)
Evaluation performance: exceeds standard Gemini 3 Flash in long-context and agent coding evaluations
Gemini 3.5 Flash Cyber: a fine-tuned model for network security scenarios
Gemini 3.5 Flash Cyber is a model fine-tuned based on 3.5 Flash and specialized for network security defense scenarios. The model can work alongside multi-agent systems such as CodeMender to quickly generate comprehensive reports, helping defenders discover and fix cybersecurity vulnerabilities. Due to risks related to dual use of the technology, access to Gemini 3.5 Flash Cyber is currently strictly limited to government agencies and trusted partner organizations, which can access it in limited quantities through pilot programs.
Gemini 4 Pre-Training Launched: Tulsee Doshi Announcement and Gemini 3.5 Pro Testing Progress
Tulsee Doshi announced in an official blog post dated July 21, 2026 that Google has officially started Gemini 4 pre-training work and described it as Google’s “most ambitious” project. Additionally, Gemini 3.5 Pro is currently undergoing testing with partner organizations; after Gemini 3.5 Pro is broadly rolled out, subsequent progress on Gemini 4 will also be announced in due course. Specific timing will be subject to Google’s official announcements.
Developers can currently access Gemini 3.6 Flash and Gemini 3.5 Flash-Lite via channels such as the Gemini API.
Frequently Asked Questions
What are the pricing and major evaluation results for Gemini 3.6 Flash?
According to Google’s official blog, Gemini 3.6 Flash is priced at $1.50 per 1 million tokens for input and $7.50 per 1 million tokens for output. In computer use capability evaluation (OSWorld-Verified), it reached 83.0%, and its output token usage is 17% lower than its predecessor (up to a 65% reduction in the DeepSWE benchmark test).
What are the output speed and pricing for Gemini 3.5 Flash-Lite?
Gemini 3.5 Flash-Lite has an output speed of 350 tokens per second. Pricing is $0.30 per 1 million tokens for input and $2.50 per 1 million tokens for output. It supports adjustable thinking levels and is suitable for low-latency tasks such as agent search and document processing.
What are the access requirements for Gemini 3.5 Flash Cyber?
Gemini 3.5 Flash Cyber is currently limited to government agencies and trusted partner organizations for limited access through pilot programs, because Google considers that such network security-related technologies carry dual-use risks. The specific process for access requests will be subject to Google’s official announcements.
