Google launched Gemini 3.5 Flash at I/O 2026, positioning it as the fastest frontier-class AI model currently available to the public. According to Google’s official announcement, the model generates output tokens four times faster than competing frontier models, while posting benchmark scores that beat Gemini 3.1 Pro on coding and multi-step agentic tasks.
For anyone who uses the Gemini app, AI Mode in Google Search, or builds with the Gemini API, this is an upgrade that arrives automatically, not behind a paywall or a settings toggle.
The speed gap matters more than it sounds. Most AI tools today ask you to trade quality for latency. Gemini 3.5 Flash is Google’s argument that you no longer have to make that choice.
TL;DR: Gemini 3.5 Flash launched May 19, 2026 at Google I/O. It is 4x faster than rival frontier models on output speed and outperforms Gemini 3.1 Pro on coding and agentic benchmarks. It is now the default model in the Gemini app and Google Search AI Mode globally. A new personal AI agent called Gemini Spark, built on 3.5 Flash, is rolling out to trusted testers now and arrives for Google AI Ultra subscribers in the US next week. Gemini 3.5 Pro is in internal testing and arrives next month.
What Gemini 3.5 Flash actually scores on the benchmarks that matter
Gemini 3.5 Flash is Google’s strongest model yet on the benchmarks that measure real-world agentic capability, not just trivia recall. On Terminal-Bench 2.1, which tests how well a model can operate like a developer at a command line over a long session, it scores 76.2%. On GDPval-AA, an evaluation of agent decision-making quality, it reaches 1656 Elo. On MCP Atlas, a test of tool-calling and multi-step coordination, it scores 83.6%.
For multimodal reasoning, specifically reading and interpreting charts and figures rather than just images, it scores 84.2% on CharXiv Reasoning.
| Benchmark | What It Tests | Gemini 3.5 Flash Score |
|---|---|---|
| Terminal-Bench 2.1 | Long-horizon developer tasks at command line | 76.2% |
| GDPval-AA | Agent decision quality (Elo rating) | 1656 Elo |
| MCP Atlas | Multi-step tool calling and coordination | 83.6% |
| CharXiv Reasoning | Chart and figure interpretation | 84.2% |
| Output speed vs rivals | Tokens per second vs other frontier models | 4x faster |
According to the Artificial Analysis index, 3.5 Flash now sits in the top-right quadrant, which maps frontier-level intelligence against high output speed. No other model currently occupies that position at this scale.
What the speed advantage actually changes for you
A 4x output speed gain is not the kind of thing you notice on a one-sentence prompt. You notice it when the model is doing a lot of work at once: building a page, reasoning through a long document, running multiple sub-tasks in sequence.
If you have used the Gemini app to draft something complex and watched it think for several seconds before producing anything, that pause gets significantly shorter.
For developers using the API, lower latency per token means agentic workflows that previously required careful batching to stay within response time budgets become considerably less fragile.
Google’s enterprise partners are already running workflows on 3.5 Flash that would not have been practical at slower speeds.
Macquarie Bank is piloting it to process 100-plus page customer documents and generate structured recommendations with low latency. Shopify is using it to run subagents in parallel for merchant growth forecasting across global markets.
Salesforce is integrating it into Agentforce to handle multi-turn, multi-subagent enterprise automation. The pattern across all of them is the same: tasks that previously took days or weeks are running in a fraction of the time, often at less than half the cost of other frontier models according to Google’s announcement.
Gemini Spark is the part most people will miss in the launch noise
Buried beneath the benchmark tables in Google’s announcement is the debut of Gemini Spark, a personal AI agent built on 3.5 Flash that runs in the background on your behalf. Google describes it as operating 24/7, taking actions on your behalf while staying under your direction.
The rollout is staged. Trusted testers get access starting May 19. Google AI Ultra subscribers in the US get the beta next week. Everyone else waits.
What Gemini Spark actually does in practice is not fully detailed yet. Google’s framing suggests it is closer to an ambient agent than a chat interface, something that monitors context and acts without being explicitly prompted every time.
That is a meaningfully different product from asking Gemini a question and reading its answer. Whether the execution matches the framing will depend on what trusted tester reports say over the coming days.
What is confirmed: Gemini Spark is powered by 3.5 Flash, which is why the speed advantage matters at the agent layer. An agent that is waiting for slow model responses adds friction to every action it takes. A 4x faster model means the gap between the agent deciding something and the agent doing something gets smaller.
Where Gemini 3.5 Pro fits and when to expect it
Google confirmed that Gemini 3.5 Pro is already in internal testing and is planned to roll out next month. No benchmark numbers for 3.5 Pro were shared in the I/O announcement, which is consistent with how Google has handled staged launches before. Flash first, Pro when it is ready.
Based on the naming pattern established with Gemini 3.1, where Pro handled tasks requiring deeper reasoning and larger context windows while Flash handled speed-sensitive workloads, 3.5 Pro will likely target the use cases where quality ceiling matters more than throughput.
Long-form document analysis, complex multi-domain reasoning, and high-stakes enterprise decisions are the obvious fits. For the AI writing comparison crowd watching how these models handle extended prose and nuanced instruction-following, 3.5 Pro will be the one to watch.
Frequently Asked Questions
Is Gemini 3.5 Flash available to me right now?
Gemini 3.5 Flash is available globally as of May 19, 2026. It is now the default model in the Gemini app and AI Mode in Google Search, so most users already have it without any action needed.
Do I need a paid plan to use Gemini 3.5 Flash?
Gemini 3.5 Flash is available at no cost through the Gemini app and Google Search AI Mode. Developers can access it via the Gemini API in Google AI Studio. Gemini Spark, the personal agent built on 3.5 Flash, requires Google AI Ultra for the beta and is US-only at launch.
What is Gemini Spark and when does it arrive?
Gemini Spark is a personal AI agent powered by Gemini 3.5 Flash that runs in the background and takes actions on your behalf. It is rolling out to trusted testers on May 19, with the beta reaching Google AI Ultra subscribers in the US the following week.
When is Gemini 3.5 Pro coming out?
Gemini 3.5 Pro is in internal testing at Google and is planned to roll out next month. No specific date has been confirmed.
How does Gemini 3.5 Flash compare to GPT and Claude?
According to Google, Gemini 3.5 Flash generates tokens four times faster than other frontier models. On agentic and coding benchmarks like Terminal-Bench 2.1 and MCP Atlas, it outperforms Gemini 3.1 Pro, which previously led comparisons against GPT-class and Claude-class models in those categories.
What to watch for in the coming weeks
Three things are worth tracking. First, independent benchmark verification, since Google’s own charts are not a neutral source and third-party evaluations from outlets like Artificial Analysis take a few days to publish after a launch.
Second, Gemini Spark trusted tester reports, which will be the first real signal of whether the ambient agent concept works in daily use or stays a demo.
Third, the 3.5 Pro release timeline, which Google has pegged to next month but has not locked to a specific date. The Pro tier is where the comparison against GPT-5 and Claude Opus becomes genuinely meaningful at the capability ceiling.
