Gemini 3.5 Flash

Overview

Gemini 3.5 Flash is the first model in Google's Gemini 3.5 family, released on May 19, 2026 as a generally available model with the API identifier gemini-3.5-flash. Google positions it as its most intelligent Flash model to date, optimized for long-horizon agentic workflows and text/code-driven interactive UI generation rather than the cheapest possible per-token cost. The model supports multimodal inputs but outputs text only; native image generation, audio generation, Live API, and Computer Use are not supported.

The model launched alongside Gemini Spark, a new personal AI agent powered by 3.5 Flash. Spark is rolling out first to trusted testers, with a U.S. Google AI Ultra beta planned next week, while Gemini 3.5 Flash itself is available from day one in the Gemini app and developer channels. A larger sibling, Gemini 3.5 Pro, has been announced as in development; Google says it is already being used internally and is planned to roll out next month. Together with Gemini 3.1 Flash-Lite, it gives developers a wider price–performance spread inside the Gemini family.

What's New

Agentic frontier performance in a Flash model

Gemini 3.5 Flash beats Gemini 3.1 Pro on Google's own agentic benchmarks — 76.2% on Terminal-Bench 2.1, 83.6% on MCP Atlas, and 1656 Elo on GDPval-AA. Google describes it as "ideal for tackling long-horizon agentic tasks," a workload that previously required a Pro-tier model. The trade-off is price: it is materially more expensive than Flash-Lite, but Google's framing is that it delivers frontier capability "at less than half the cost of other frontier models" in its competitive class.

Stronger multimodal UI and graphics generation

Multimodal understanding is upgraded for interactive, code-rendered output, with 84.2% on CharXiv Reasoning. Google highlights that 3.5 Flash "generates richer, more interactive web UIs and graphics," making it a meaningful step up from earlier Gemini 3 Flash variants for teams building UI prototypes, in-app visualizations, or agent surfaces that ship rendered output.

4X faster output than other frontier models

Google attributes a 4X faster output tokens-per-second figure to 3.5 Flash versus other frontier models in its competitive set. Combined with the new Batch and Flex modes (both at 50% off Standard pricing), this targets throughput-bound and offline agentic pipelines.

Powers Gemini Spark and ships in Antigravity

Gemini Spark, Google's new personal AI agent for everyday digital tasks, runs on 3.5 Flash. The model is also generally available through Google Antigravity for agentic development and in the consumer Gemini app, in addition to the Gemini API in Google AI Studio, Vertex AI, Android Studio, Gemini Enterprise Agent Platform, Gemini Enterprise, and AI Mode in Search.

Performance Benchmarks

Numbers come from Google's May 19, 2026 announcement. Google states 3.5 Flash outperforms Gemini 3.1 Pro on the coding and agentic benchmarks listed above; per-task results will vary by workload, prompt strategy, and token mix.

Migration Guide

From Gemini 3.1 Flash-Lite

3.5 Flash is not a drop-in cost-equivalent upgrade. Standard pricing moves from $0.25/$1.50 per 1M input/output tokens (Flash-Lite) to $1.50/$9.00 — roughly a 6× increase — in exchange for stronger agentic and multimodal performance. Recommended path:

• Keep Flash-Lite for high-volume, latency-sensitive, cost-bounded tasks (translation, classification, structured extraction, lightweight chat).
• Move to 3.5 Flash only for tasks where agentic quality or richer multimodal output justifies the higher per-token cost.
• Re-run cost-per-task and quality evals on your own data before migrating production traffic.

From Gemini 3.1 Pro (Preview)

For teams using gemini-3.1-pro-preview for agentic or coding workflows, 3.5 Flash is a GA stable alternative that posts higher scores on Terminal-Bench 2.1, MCP Atlas, GDPval-AA, and CharXiv Reasoning per Google's announcement. Update the model identifier to gemini-3.5-flash, validate tool-use and structured-output behavior on your suite, and consider whether the pending Gemini 3.5 Pro (announced for the following month) is a better fit for your heaviest reasoning workloads.

API and channel changes

The model is available via the Gemini API in Google AI Studio, Vertex AI, and Android Studio, in Google Antigravity for agentic development, in Gemini Enterprise Agent Platform, Gemini Enterprise, AI Mode in Search, and inside the Gemini app for end users. Free-tier usage is available; production deployments should pick a paid inference mode (Standard, Batch, Flex, or Priority) based on latency and budget. Developers migrating from Gemini 3 Flash Preview should also update gemini-3-flash-preview to gemini-3.5-flash, replace thinking_budget with thinking_level, add id and matching name to FunctionResponse parts, account for the default effort change from high to medium, and avoid Computer Use workloads on 3.5 Flash.

Pricing & Plans

Gemini consumer access via the Gemini app remains free for general use. The developer model is billed per token through the Gemini API with a free tier and four paid inference modes.

Batch and Flex both deliver a 50% discount versus Standard at the cost of higher latency or relaxed scheduling. Priority is roughly 1.8× Standard for guaranteed capacity. Gemini 3.5 Flash uses a unified paid input price in the Gemini API pricing table; its supported inputs include text, image, video, audio, and PDF. Confirm current rates on Google's official Gemini API pricing page before committing production budgets.

Best For

• Long-horizon agentic workflows (multi-step planning, tool use, code editing) where 3.1 Flash-Lite was too weak and 3.1 Pro was too expensive
• Teams building or running production agents on Google Antigravity
• Multimodal applications that generate text/code-driven interactive web UIs, dashboards, or rendered in-app graphics
• Throughput-bound offline pipelines that can use Batch or Flex at a 50% discount
• Product teams previously evaluating Claude or GPT frontier models that now want a Gemini-native option

FAQ