Google Launches New Gemini AI Models While Pro Version Delayed

On July 21, 2026, Google announced the launch of new AI models, including the Gemini 3.6 Flash and 3.5 Flash-Lite, overshooting the expected rollout of the Gemini 3.5 Pro. The 3.6 Flash model boasts a 17% reduction in output token usage—output tokens serving as a fundamental processing unit for AI, approximately equivalent to three-quarters of a word—bringing the operational cost down from $9 to $7.50 per million tokens.

Alphabet Inc.'s share price witnessed a notable decline of about 4.4%, amounting to a loss of roughly $200 billion in market capitalization, following news that the 3.5 Pro model did not meet internal standards particularly regarding coding tasks. A planned fix involving updated training datasets failed to deliver the expected improvements. This delay comes after the last Pro-model, Gemini 3.1 Pro, was launched in February 2026.

The flash series models are designed for efficiency, focusing on speed and cost-effectiveness for AI agents—automated systems that can execute tasks like data processing and managing documents without requiring human intervention. According to the Artificial Analysis Index, the Gemini 3.6 Flash model achieved a 49% completion rate on the DeepSWE v1.1 benchmark, which evaluates software engineering tasks.

In contrast, Pro models are characterized by higher computational power, aimed at complex reasoning tasks, albeit at a slower and pricier rate. As Google progresses towards the next version, the tech giant announced the initiation of pre-training for Gemini 4, which they describe as "our most ambitious pre-training run yet." This dual release strategy of the new Flash models while postponing the Pro version reflects a cautious approach amid fluctuating market conditions.

The future of Google's AI offerings will depend on successfully navigating the performance targets set for the upcoming Pro model, as well as the broader market response to the newly launched Flash models.