Google launched the Gemini 3.6 Flash and 3.5 Flash-Lite AI models but delayed the Pro version due to performance issues.
Summary
- Google launched the Gemini 3.6 Flash and 3.5 Flash-Lite models, boasting enhanced efficiency and reduced costs, while the Gemini 3.5 Pro version remains delayed due to unresolved internal issues.
- The Gemini 3.6 Flash model requires 17% fewer output tokens than its predecessor, lowering costs from $9 to $7.50 per million tokens, significantly benefitting AI operations.
- Additionally, Google has begun pre-training for Gemini 4, marking it as the company's most ambitious AI initiative to date.
Today, Google announced the release of the Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber models, although the anticipated Pro version was notably absent.
The delay of the Gemini 3.5 Pro, initially expected shortly after the May introduction of 3.5 Flash, stems from it not meeting internal performance benchmarks. This setback led to a significant drop in Alphabet’s stock, which fell by approximately 4.4%, equating to a $200 billion loss in market value.
The last Pro model released was Gemini 3.1 Pro in February, and the new Flash series aims to deliver quick, cost-effective solutions for AI tasks like document management and data analysis. In contrast, Pro models are intended for more complex tasks but are slower and pricier.
Details on New AI Models
The standout release, Gemini 3.6 Flash, uses 17% fewer output tokens, which represent the basic units processed by AI. The new model costs $1.50 per million input tokens and $7.50 per million output tokens, a notable decrease from the previous model's output cost. This reduction is crucial for businesses deploying AI at scale.
In performance benchmarks, 3.6 Flash achieved a score of 49% on DeepSWE v1.1, surpassing the previous version's score of 37%. It also scored 63.9% on MLE-Bench, up from 49.7%, and excelled in OSWorld-Verified tests, scoring 83.0%, outpacing competitors like Claude Sonnet 5 and GPT-5.6 Luna.
However, the model faced challenges in coding tasks during tests, producing unsatisfactory results initially. After troubleshooting with Deepseek, several critical fixes were made, leading to improved performance.
The Gemini 3.5 Flash-Lite model is optimized for high-volume tasks, processing 350 output tokens per second at a low cost of $0.30 per million input tokens and $2.50 for output tokens, making it suitable for large-scale document processing and agentic search systems.
Additionally, the Gemini 3.5 Flash Cyber model will be restricted to government and trusted partners for addressing software vulnerabilities, as Google is cautious about broader access to this capability.
Meanwhile, Google's DeepMind team has commenced pre-training for Gemini 4, which is described as its most ambitious project, indicating progress beyond mere planning stages.
We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress : )
— Logan Kilpatrick (@OfficialLoganK) July 21, 2026
Both the 3.6 Flash and 3.5 Flash-Lite models are currently available in the Gemini app, Google AI Studio, and through the API. Google has stated that the Gemini 3.5 Pro will be released once it meets performance criteria, although a timeline remains uncertain.
