Google: Gemini 3.5 Flash

Gemini 3.5 Flash is the first model in Google's Gemini 3.5 family, announced at Google I/O on May 19, 2026. Google describes the release as combining "frontier intelligence with action": it is engineered first and foremost for agents and coding, not chat. According to DeepMind, 3.5 Flash outperforms Gemini 3.1 Pro, the previous frontier model, on nearly all benchmarks while keeping the low latency and cost profile of the Flash line.

Key Features

• Agent-first architecture: 3.5 Flash excels at complex, long-horizon tasks with sustained frontier performance, including sub-agent deployment, rapid agentic loops, and multi-step tool use at scale.
• 1M token context: Supports a 1 million token input context window with up to 65k output tokens, enough for entire codebases or long research pipelines.
• Thinking levels: Default thinking effort moved from high to medium, and the low effort setting was significantly improved for coding and agentic tasks that need fewer steps at lower latency and cost.
• Thought preservation: The model maintains intermediate reasoning across multi-turn conversations automatically, with no API changes needed.
• Multimodal foundation: Builds on Gemini 3's strong multimodal understanding, with leading scores on vision-language benchmarks such as CharXiv (84.2%).

Performance Highlights

• Terminal-Bench 2.1: 76.2% on the agentic coding evaluation
• GDPval-AA: 1656 Elo on economically valuable knowledge-work tasks
• MCP Atlas: 83.6%, leading among evaluated models
• CharXiv: 84.2% on multimodal understanding
• Runs at less than half the cost of comparable models for long-horizon agentic workloads

Availability & Pricing

Gemini 3.5 Flash is generally available via the Gemini API, Google AI Studio, Android Studio, Google Antigravity, and Gemini Enterprise. It is also the default model for the Gemini app and AI Mode in Search globally. Pricing is $1.50 per million input tokens and $9.00 per million output tokens (including thinking tokens), with cached input at $0.15 per million. A rate-limited free tier is available in AI Studio.

At release, Gemini 3.5 Pro was already in internal testing, with a public rollout planned for the following month.

Use Cases

• Autonomous coding agents: The model can execute multi-step coding pipelines, explore alternate paths, and prototype rapidly, which Google demonstrated at I/O by building a working operating system from scratch.
• Agent orchestration at scale: With Antigravity, 3.5 Flash coordinates sub-agent teams for demanding production workloads.
• Long-horizon research: Its 1M context and thought preservation make it well suited to research projects that span many steps and large amounts of source material.
• Cost-sensitive production: The Flash pricing tier and improved low-effort mode make it a practical default for high-volume agentic APIs.

Getting Started

1. Open Google AI Studio and select gemini-3.5-flash in the model picker.
2. Generate an API key from Google Cloud Console.
3. Call the Gemini API with the model ID gemini-3.5-flash; input is multimodal (text, images, audio, video).
4. Tune thinking effort (low/medium/high) per request to balance quality, cost, and latency.

Advantages & Comparison

Compared to the Gemini 3 Flash it replaces, 3.5 Flash delivers frontier-level agentic and coding performance rather than mid-tier capability, at a modest price increase (migration from gemini-3-flash-preview costs more but buys GA stability and substantially stronger reasoning). Compared to other frontier models, it combines near-flagship intelligence with the speed and price of a Flash-tier model, which is unusual in the market.

Frequently Asked Questions

What is the knowledge cutoff?

Gemini 3.5 Flash has a knowledge cutoff of January 2025. For current information, use the Search Grounding tool.

Is 3.5 Flash open source?

No. Like the rest of the Gemini family, it is proprietary and available only through Google's products and API.

When is Gemini 3.5 Pro coming?

Google was using 3.5 Pro internally at launch and planned a public rollout roughly one month after the I/O release.

Alternatives

• Claude Opus 5: Anthropic's frontier model, with strong agentic coding and a 1M-token context: /en/items/models/claude-opus-5
• GPT-5.6: OpenAI's flagship reasoning family: /en/items/models/gpt-5.6
• GLM-5.3: Zhipu's open-weight coding and agentic flagship: /en/items/models/glm-5-3

Conclusion

Gemini 3.5 Flash marks Google's decisive bet that the next AI wave is agentic, not conversational. It pairs near-frontier intelligence with Flash-tier speed and cost, making long-horizon agentic workloads practical at scale. Developers building autonomous agents, coding pipelines, or research automation should treat it as a default candidate.

Usage Terms

Usage is subject to Google's Gemini API Terms of Use. Note that data sent through the free tier may be used to improve Google products.