Gemini 3.6 Flash
Gemini 3.6 Flash is the Flash-tier workhorse of the Gemini 3.x line, improving coding, agentic execution, and web development output while consuming fewer output tokens and making fewer model calls than Gemini 3.5 Flash, with built-in computer use and a context window of 1M tokens. Your use is subject to Google's Terms & Privacy Policies.
import { streamText } from 'ai'
const result = streamText({ model: 'google/gemini-3.6-flash', prompt: 'Why is the sky blue?'})Playground
Try out Gemini 3.6 Flash by Google. Usage is billed to your team at API rates. Free users (those who haven't made a payment) get $5 of credits every 30 days.
Gemini 3.6 Flash
Gemini 3.6 Flash
Providers
Route requests across multiple providers. Copy a provider slug to set your preference. Visit the docs for more info. Using a provider means you agree to their terms, listed under Legal.
P50 throughput on live AI Gateway traffic, in tokens per second (TPS). Visit the docs for more info.
P50 time to first token (TTFT) on live AI Gateway traffic, in milliseconds. View the docs for more info.
Direct request success rate on AI Gateway and per-provider. Visit the docs for more info.
More models by Google
| Model |
|---|
| 1M | 0.5s | 348tps | $0.30/M | $2.50/M | Read:$0.03/M Write:— | $14/K+1 more + input costs | +3 |  | 07/21/2026 | ||||
| 1M | 1.9s | 150tps | $1.50/M | $9/M | Read:$0.15/M Write:— | $14/K+1 more + input costs | +3 | | 05/19/2026 | ||||
| 1M | 0.4s | 309tps | $0.25/M | $1.50/M | Read:$0.03/M Write:— | $14/K+1 more + input costs | +3 | | 05/07/2026 | ||||
| 1M | 2.2s | 103tps | $2/M+1 more | $12/M+1 more | Read: $0.2/M+1 more Write: — | $14/K+1 more + input costs | +3 | | 02/19/2026 | ||||
| 1M | 0.7s | 181tps | $0.50/M+1 more | $3/M+1 more | Read: $0.05/M+1 more Write: — | $14/K+1 more + input costs | +3 | | 12/17/2025 | ||||
| 1M | 0.2s | 405tps | $0.10/M | $0.40/M | Read:$0.01/M Write:— | $35/K+1 more + input costs | +3 | | 06/17/2025 |
About Gemini 3.6 Flash
Gemini 3.6 Flash is Google's Flash-tier workhorse, released on July 21, 2026, and it builds directly on Gemini 3.5 Flash. Quality rises across coding, knowledge work, and multimodal tasks while token consumption falls. On the Artificial Analysis Index, Gemini 3.6 Flash consumes fewer output tokens than Gemini 3.5 Flash, and it completes multi-step workflows with fewer reasoning steps and fewer tool calls. For agent workloads, that compounds: the saving lands on every step of a run, not on a single call.
Coding is where the gains are easiest to measure. Gemini 3.6 Flash scores 49% on DeepSWE against 37% for Gemini 3.5 Flash, and 63.9% on MLE Bench against 49.7%. Generated code arrives with fewer unwanted edits and less execution looping, and web and app development output comes back cleaner, which matters when an agent writes a multi-element layout you intend to ship rather than rewrite.
Computer use ships as a built-in client-side tool. Gemini 3.6 Flash scores 83.0% on OSWorld-Verified against 78.4% for Gemini 3.5 Flash, so agents that drive a browser or a desktop application close the gap between a plan and a completed action more often. Multimodal and spatial reasoning improved alongside it, covering chart interpretation, visual blueprint conversion, and multi-element web layout generation. Gemini 3.6 Flash accepts text, images, audio, video, and PDF documents within a context window of 1M tokens and returns text, up to 64K tokens.
Knowledge work moved with the rest: Gemini 3.6 Flash scores 1421 on GDPval-AA v2 against 1349 for Gemini 3.5 Flash, and Google points to document parsing, chart and data analysis, and report drafting as the tasks customers exercise most. Function calling, structured outputs, code execution, search grounding, URL context, and context caching are all supported. Calling Gemini 3.6 Flash through AI Gateway adds usage and cost tracking, automatic retries, and provider failover on one API surface, with no Google Cloud account required.
What To Consider When Choosing a Provider
- Configuration: Gemini 3.6 Flash defaults to the
mediumthinking level. SetthinkingLevelunderproviderOptions.google.thinkingConfigtominimalorlowfor extraction, routing, and classification, and tohighwhen an agent has to reason across many steps. Thinking tokens count toward output tokens, so the level you pick moves the cost of every request. Budget against realistic thinking settings rather than the default before you size a production deployment. - Zero Data Retention: AI Gateway supports Zero Data Retention for this model via direct gateway requests (BYOK is not included). To configure this, check the documentation.
- Authentication: AI Gateway authenticates requests using an API key or OIDC token. You do not need to manage provider credentials directly.
When to Use Gemini 3.6 Flash
Best for
- Agentic Coding Loops: Fewer reasoning steps and tool calls per task cut the cost of a full multi-step run
- Production Code Generation: DeepSWE and MLE Bench results improve over Gemini 3.5 Flash with fewer unwanted edits
- Computer Use Agents: A built-in computer use tool and higher OSWorld-Verified results support browser and desktop automation
- Web and App Development: Cleaner layout and component output leaves less generated frontend code to rewrite
- Multimodal Knowledge Work: Chart interpretation, document parsing, and report drafting all improved in this release
Consider alternatives when
- Deepest Reasoning Required:
google/gemini-3.1-pro-previewcarries more depth on hard multi-step problems - Volume Over Capability:
google/gemini-3.5-flash-literuns high-throughput extraction and classification more cheaply - Native Image Output:
google/gemini-3.1-flash-imageandgoogle/gemini-3-pro-imagereturn images; Gemini 3.6 Flash returns text - Video Generation Work:
google/veo-3.1-generate-001andgoogle/gemini-omni-flash-previewproduce video output
Conclusion
Gemini 3.6 Flash is the Flash-tier default for agentic coding and multimodal knowledge work on AI Gateway. It posts better coding, computer use, and knowledge work results than Gemini 3.5 Flash while spending fewer output tokens and fewer tool calls per task, which lowers the cost of a whole agent run rather than a single call.