Gemini 3.5 Flash

Gemini 3.5 Flash is Google DeepMind's most capable Flash-tier model, unveiled at Google I/O on May 19, 2026. Positioned as the fastest frontier AI model available, it targets the growing demand for AI agents that can autonomously execute complex, multi-step tasks in software engineering, data analysis, and business automation.

The model's headline metric is speed: 289 tokens per second output, making it approximately four times faster than other frontier models. This speed doesn't come at the expense of quality — Gemini 3.5 Flash scores within two points of Anthropic's flagship model on major coding benchmarks while costing roughly one-third the price. On Terminal-Bench 2.1, a rigorous software engineering benchmark, it achieves 76.2%. Its GDPval-AA Elo rating of 1656 places it among the top coding models globally.

A key innovation is the multi-agent architecture with the Antigravity harness system. This enables Gemini 3.5 Flash to spawn and coordinate parallel subagents for concurrent task execution. In practice, this means the model can simultaneously research documentation, write code, run tests, and review results — dramatically accelerating complex software development workflows. The architecture is purpose-built for agentic scenarios where the model must plan, execute, observe results, and iterate autonomously.

The model excels in tool use scenarios, scoring 83.6% on MCP Atlas, which evaluates how effectively models interact with external tools and APIs. Combined with its 84.2% on CharXiv Reasoning for multimodal understanding, it can process technical diagrams, charts, and visual documentation alongside code — a critical capability for real-world engineering work.

With a 1 million token context window, Gemini 3.5 Flash can ingest entire codebases, lengthy technical specifications, and extended conversation histories without losing coherence. This is particularly valuable for code understanding tasks where the model needs to reason about the relationships between files spread across a large project.

Compared to its predecessor Gemini 2.5 Flash, the 3.5 version shows 10-20% improvement in low-reasoning coding performance, significantly better handling of long-horizon tasks, and enhanced multi-agent coordination. It also produces richer, more interactive web UIs and graphics when given generation tasks.

Pricing follows Google's aggressive Flash-tier strategy: $1.50 per million input tokens and $9.00 per million output tokens — roughly 5x cheaper than input and 3.6x cheaper than output compared to frontier Pro-tier models. The 1M token context window is included at no additional premium. This pricing makes it economically viable for always-on agent scenarios that would be prohibitively expensive with other frontier models.

Gemini 3.5 Flash is generally available from launch through Google AI Studio and the Gemini API, with integration support for all major AI development frameworks. The 3.5 Pro variant is in internal testing with public release planned for the following month.