Google unveils three Gemini models but delays the highly anticipated Pro version

⚡

Key Takeaways

1Google has introduced the Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber models, each with distinct features.

2The 3.6 Flash model stands out with a 17% reduction in token usage, thereby improving its efficiency.

3The 3.5 Flash Cyber model, focused on cybersecurity, is reserved for government partners in a pilot program.

💡Why it matters — Google is looking to strengthen its position against the intense competition from OpenAI and Anthropic with these new models, while delaying the update of its flagship model.

Google Unveils Its New Gemini Models

This Tuesday, Google DeepMind announced the launch of three new artificial intelligence models: Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. The 3.6 Flash model positions itself as Google's flagship model, offering significant improvements in areas such as programming, knowledge work, and multimodal performance. One notable feature of this model is its ability to reduce token usage by 17%, making it more cost-effective than the previous model, the 3.5 Flash.

Models Tailored to Specific Needs

The Gemini 3.5 Flash-Lite model is designed to be the most cost-effective in the lineup, while the 3.5 Flash Cyber specializes in detecting and correcting cybersecurity vulnerabilities. The latter will be offered exclusively to governments and trusted partners as part of a restricted access pilot program, as specified by Google.

Performance and Reliability Goals

Google emphasized that these new models aim to provide increased efficiency, latency, and reliability for clients developing large-scale AI agents. The launch of these models is significant not only for the innovations they bring but also for what is missing: the anticipated update of the flagship model, Gemini Pro.

Notable Absence of Gemini Pro

While Google has delivered faster and less expensive models optimized for programming, efficiency, and cybersecurity, the absence of the update for Gemini Pro is striking. This model has not been updated since February, and its absence contrasts with the recent launches from OpenAI and Anthropic, which introduced GPT-5.5, GPT-5.6, Claude Opus 4.8, and Claude Sonnet 5.

Delays in the Launch of Gemini Pro

Google had initially planned to launch the Pro version in May, stating that it was already in use internally and would be deployed the following month. However, Bloomberg recently reported that Google is facing internal delays in meeting the performance targets set for the 3.5 Pro.

On the Road to Gemini 4

Gemini Pro models are generally Google's highest-performing models, designed for complex reasoning and programming tasks, while Flash models are optimized for lower cost and fast response times. Logan Kilpatrick, product lead at Google DeepMind, stated that the company is currently testing Gemini 3.5 Pro with partners and hopes to deploy it soon. He also mentioned that the team has begun its most ambitious pre-training run for Gemini 4.