Google Gemini 3.5 Flash: The New AI Benchmark Surpasses Its Rivals

Google Unveils Gemini 3.5 Flash

On May 19, 2026, at the Google I/O event, Google unveiled Gemini 3.5 Flash, its latest artificial intelligence model. This model stands out for its speed, reaching up to 300 tokens per second, making it a top choice for autonomous AI agents. Gemini 3.5 Flash is also one of the fastest and most cost-effective versions in Google's lineup.

Exceptional Performance

Gemini 3.5 Flash outperforms its predecessor, Gemini 3.1 Pro, across most agent and coding benchmarks. In terms of speed, it is designed to be four times faster than other leading AIs, making it particularly suited for complex and lengthy tasks. Furthermore, it competes with Claude Opus 4.7, Anthropic's most powerful model, on certain criteria.

Comparison with Competing Models

In comparative tests, Gemini 3.5 Flash demonstrated impressive performance:

• Terminal-Bench 2.1: 76.2% for Gemini 3.5 Flash, compared to 70.3% for Gemini 3.1 Pro and 78.2% for GPT-5.5.
• GDPval-AA Elo: 1656 for Gemini 3.5 Flash, surpassing the 1204 of Gemini 3.1 Pro, and approaching the 1753 of Opus 4.7 and the 1769 of GPT-5.5.
• MCP Atlas: 83.6% for Gemini 3.5 Flash, compared to 78.2% for Gemini 3.1 Pro, 79.1% for Opus 4.7, and 75.3% for GPT-5.5.
• CharXiv Reasoning: 84.2% for Gemini 3.5 Flash, slightly above the 83.3% of Gemini 3.1 Pro, the 82.1% of Claude Opus 4.7, and the 84.1% of GPT-5.5.

Applications and Availability

Google emphasizes that Gemini 3.5 Flash is designed to reduce the workload of developers and auditors by enabling them to plan, develop, and iterate to solve real-world problems. It can also orchestrate multiple AI sub-agents to manage large-scale tasks.

This model is now accessible via Google AI Studio, Android Studio, the Google Antigravity platform, as well as on Gemini Enterprise and the Gemini app. However, it is not available in France, although it is being deployed globally as the default model in the search engine and AI mode.