Gemini 3.6 Flash: Benchmarks, Pricing, Availability, and What Changed
Google released Gemini 3.6 Flash on July 21, 2026 as the new general-purpose workhorse in its Flash lineup. The model targets coding, multimodal analysis, computer-use agents and multi-step tool workflows, while reducing output-token consumption and lowering the output-token price compared with Gemini 3.5 Flash.
The headline figures are substantial: Google reports 49% on DeepSWE, 63.9% on MLE-Bench and 83.0% on OSWorld-Verified. Standard API pricing is $1.50 per million input tokens and $7.50 per million output tokens. This guide explains the benchmarks, pricing tiers, availability, API changes, supported tools and the practical reasons to migrate.
Source: blog.google
Key takeaways
- Release: Gemini 3.6 Flash became generally available on July 21, 2026.
- Model ID: The stable Gemini API model code is
gemini-3.6-flash. - Pricing: Standard pricing is $1.50 per million input tokens and $7.50 per million output tokens.
- Efficiency: Google reports 17% fewer output tokens on the Artificial Analysis Index and up to 65% fewer on DeepSWE than Gemini 3.5 Flash.
- Context: The model accepts up to 1,048,576 input tokens and produces up to 65,536 output tokens.
- Availability: It is available through the Gemini API, Google AI Studio, Android Studio, Google Antigravity, Gemini Enterprise and the Gemini app.
Gemini 3.6 Flash at a glance
| Category | Gemini 3.6 Flash |
|---|---|
| Primary role | Fast general-purpose reasoning model for agentic and multimodal workloads |
| Stable API identifier | gemini-3.6-flash |
| Default thinking level | Medium |
| Input modalities | Text, image, video, audio and PDF |
| Output modality | Text |
| Production status | Generally available |
What is Gemini 3.6 Flash?
Gemini 3.6 Flash is a proprietary multimodal reasoning model in Google’s speed-and-efficiency-focused Flash family. Google positions it for real-world agentic workloads: code generation, function calling, computer use, spatial reasoning, document analysis and workflows that may require several steps before a final response is ready.
The model is not merely a renamed 3.5 release. Google says it was tuned around developer feedback concerning verbosity, unwanted code edits and execution loops that consumed additional reasoning tokens and tool calls. The intended improvement is therefore twofold: better task quality and less work required to reach the result.
For developers starting with Google’s platform, Zerlo’s guide to obtaining a Gemini API key explains the account and Google AI Studio setup. The model name to select in an API request is gemini-3.6-flash.
What changed from Gemini 3.5 Flash?
| Area | Gemini 3.6 Flash change | Practical effect |
|---|---|---|
| Coding | Higher precision, fewer unintended file changes and fewer debugging loops | Better suited to migrations and iterative agentic development |
| Token efficiency | Fewer output tokens, reasoning steps, turns and tool calls | Potentially lower cost per completed task rather than only per token |
| Multimodal reasoning | Improved chart interpretation, blueprint conversion and multi-element layout understanding | Stronger document, visual and spatial workflows |
| Computer use | Native tool support, currently marked as Preview in the model documentation | Can power supported browser and interface automation workflows |
| Output pricing | $7.50 per million tokens instead of $9.00 for Gemini 3.5 Flash | A 16.7% reduction in the standard output-token rate |
| API configuration | Sampling parameters such as temperature, top_p and top_k are deprecated |
Existing integrations need configuration review, not only a model-ID change |
There is one nuance for frontend work. Google’s migration documentation says Gemini 3.6 Flash tends to inspect a project programmatically before making changes. That can improve correctness on complex tasks, but it may add exploratory steps to a simple visual edit. Google also notes that human evaluators sometimes preferred earlier models for styling, so explicit design instructions remain valuable.
Gemini 3.6 Flash benchmarks
Google’s launch data shows gains over Gemini 3.5 Flash in coding, machine-learning engineering, knowledge work and computer-use tasks. These benchmarks use different scoring methods, so their values should not be averaged into one overall percentage. Each result is evidence for a specific type of workload.
| Benchmark | What it tests | Gemini 3.5 Flash | Gemini 3.6 Flash |
|---|---|---|---|
| DeepSWE v1.1 | Long-horizon software-engineering agents | 37% | 49% |
| MLE-Bench | Machine-learning engineering and research tasks | 49.7% | 63.9% |
| OSWorld-Verified | Computer use in real desktop environments | 78.4% | 83.0% |
| GDPval-AA v2 | Knowledge-work task quality | 1349 | 1421 |
Source: blog.google
The strongest published gains appear in long-horizon software engineering and machine-learning engineering, while computer-use performance also moves higher.
How to read these benchmark results
- DeepSWE matters for autonomous coding: it is more relevant to repository-level work than a short standalone coding question.
- MLE-Bench measures a specialist workflow: the result is useful for data-science and machine-learning agents, but not a general chatbot score.
- OSWorld-Verified tests actions: a higher score supports the computer-use story, although the related API capability is still marked Preview.
- GDPval-AA is not a percentage: it is a comparative knowledge-work evaluation and should be read within that benchmark’s methodology.
Token efficiency is part of the benchmark story
Raw task scores do not show how many tokens or tool calls were needed to reach the result. Google says Gemini 3.6 Flash generated 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Intelligence Index. On DeepSWE, the company reports a larger reduction: average output tokens per task fell from about 276,000 to 97,000, a decrease of roughly 65%.
Artificial Analysis independently lists the high-thinking configuration of Gemini 3.6 Flash with an Intelligence Index score of 50 and a one-million-token context window. Benchmark results can change as evaluators update prompts, harnesses and model versions, so production testing with representative tasks remains more important than a leaderboard position.
Source: blog.google
Reduced output-token use can lower the cost of long agent loops, especially when the model repeatedly reasons, edits files and calls tools before finishing.
Gemini 3.6 Flash pricing
Google’s standard published rate is $1.50 per million input tokens and $7.50 per million output tokens. The input rate is unchanged from Gemini 3.5 Flash, while the output rate falls from $9.00 to $7.50. Cached input is listed at $0.15 per million tokens on Google Cloud’s Agent Platform pricing page.
| Consumption option | Input per 1M tokens | Output per 1M tokens | Typical use |
|---|---|---|---|
| Standard | $1.50 | $7.50 | Normal interactive and production API traffic |
| Priority | $2.70 | $13.50 | Workloads that pay more for prioritized inference capacity |
| Flex or Batch | $0.75 | $3.75 | Delay-tolerant or asynchronous processing |
Example standard-price calculations
| Monthly usage example | Input cost | Output cost | Estimated total |
|---|---|---|---|
| 10M input and 1M output tokens | $15.00 | $7.50 | $22.50 |
| 50M input and 5M output tokens | $75.00 | $37.50 | $112.50 |
| 100M input and 20M output tokens | $150.00 | $150.00 | $300.00 |
These examples cover model tokens only. Search grounding, computer use, code execution, external services and other platform components can add cost or latency. The most useful internal metric is therefore cost per successful completed task, including retries and tool calls, rather than the listed token rate alone.
Availability: where can you use Gemini 3.6 Flash?
Gemini 3.6 Flash was available from launch day across Google’s consumer, developer and enterprise surfaces. Exact usage limits, regional access and model-selection controls can differ by product, subscription and account.
| User group | Available through | Important detail |
|---|---|---|
| Developers | Gemini API, Google AI Studio and Android Studio | Use the stable model ID gemini-3.6-flash |
| Agent developers | Google Antigravity and Gemini Managed Agents | Google made 3.6 Flash the default model for the Antigravity managed agent |
| Enterprises | Gemini Enterprise Agent Platform and the Gemini Enterprise app | Enterprise billing, governance and region settings apply |
| General users | Gemini app | Access and usage limits can depend on the account and subscription tier |
Source: blog.google
AI Studio logs and datasets can help teams compare model behavior, inspect failed requests and build repeatable evaluation sets before changing a production deployment.
Gemini 3.6 Flash should not be confused with Gemini 3.5 Flash-Lite or Gemini 3.5 Flash Cyber. Flash-Lite is the lower-cost, high-throughput option. Flash Cyber is a specialized security model planned for governments and trusted partners through a limited-access CodeMender pilot, not a generally available replacement for 3.6 Flash.
Context window, inputs and supported tools
| Specification | Gemini 3.6 Flash |
|---|---|
| Input types | Text, image, video, audio and PDF |
| Output type | Text |
| Input token limit | 1,048,576 tokens |
| Output token limit | 65,536 tokens |
| Supported capabilities | Caching, code execution, file search, function calling, search grounding, Maps grounding, structured output, thinking and URL context |
| Computer use | Supported in Preview |
| Consumption options | Standard, Batch API, Flex inference and Priority inference |
| Not supported | Image generation, audio generation and the Live API |
The million-token context window is useful for large document sets, code repositories, transcripts and multimodal records. It does not guarantee perfect recall across every position in a long prompt, and sending the maximum context can increase cost and latency. Retrieval, file search and prompt segmentation may still produce a more reliable production system.
Source: blog.google
File Search is one of the built-in tools supported by Gemini 3.6 Flash and can ground responses in indexed project documents instead of placing every file directly into each prompt.
Which built-in tools matter most?
- File Search: useful for retrieval over internal documents, manuals and knowledge bases.
- Code execution: lets the model calculate, transform data and verify parts of an answer programmatically.
- Function calling: connects the model to application-specific actions and data sources.
- Google Search and Maps grounding: adds current or location-aware information where supported.
- Computer use: enables interface automation, but remains a Preview capability and requires careful permissions and validation.
Migration checklist for Gemini 3.6 Flash
- Change the target model ID to
gemini-3.6-flashin a test environment. - Remove deprecated sampling fields such as
temperature,top_pandtop_k. - Replace older thinking-budget settings with an appropriate
thinking_level. - Remove prefilled model turns and review multi-turn state handling.
- Re-test function calls, tool responses and structured-output schemas.
- Compare task success, latency, input tokens, output tokens, turns and tool calls against the current production model.
- Use a staged rollout and keep a rollback path until representative workloads are stable.
Should you migrate from Gemini 3.5 Flash?
Gemini 3.6 Flash is the logical migration target when an application depends on coding agents, multi-step tool use, computer interaction, document analysis or visual and spatial reasoning. The lower output price and reduced tendency to loop are especially relevant for agents that may call the model repeatedly before completing one user request.
| Workload | Best current direction | Reason |
|---|---|---|
| Repository-level coding agent | Test Gemini 3.6 Flash | Improved coding precision and fewer execution loops |
| Multimodal document and chart analysis | Test Gemini 3.6 Flash | Stronger published multimodal and knowledge-work results |
| High-volume classification or extraction | Compare with Gemini 3.5 Flash-Lite | Flash-Lite is designed for lower-cost throughput |
| Image generation | Choose another Gemini model | Gemini 3.6 Flash produces text output only |
| Real-time voice application | Use a Live API-compatible model | The Live API is not supported by Gemini 3.6 Flash |
| Stable legacy pipeline with no quality issues | Benchmark before migrating | A change is useful only when quality, latency or cost improves on the actual workload |
For a broader model-selection perspective, Zerlo’s Gemini vs. Claude technical comparison explains the different trade-offs between Google’s ecosystem and Anthropic’s models. Users focused on reusable assistants rather than API deployment may also find the Gemini Gems guide useful.
FAQ
When was Gemini 3.6 Flash released?
Google announced and released Gemini 3.6 Flash on July 21, 2026. It became generally available through developer, enterprise and consumer Gemini products.
What is the Gemini 3.6 Flash API model name?
The stable model ID is gemini-3.6-flash. Google’s model documentation lists it as a stable version rather than a preview-only alias.
How much does Gemini 3.6 Flash cost?
Standard pricing is $1.50 per million input tokens and $7.50 per million output tokens. Google Cloud also lists lower Flex or Batch rates and higher Priority rates. Tools and other platform services may add separate charges.
Is Gemini 3.6 Flash cheaper than Gemini 3.5 Flash?
Yes for standard output tokens. The input price remains $1.50 per million tokens, while the output price drops from $9.00 to $7.50 per million tokens. Google also reports reduced output-token use on its evaluations.
Does Gemini 3.6 Flash support images, audio and video?
It accepts text, images, video, audio and PDF documents as input. Its output is text. It does not support image generation, audio generation or the Live API.
Is Gemini 3.6 Flash available in the Gemini app?
Yes. Google listed the Gemini app among the launch surfaces. Account-specific limits, regional rollout details and model-selection controls can still vary.
Does Gemini 3.6 Flash support computer use?
Yes, but Google marks computer use as a Preview capability in the model documentation. Production systems should limit permissions, validate actions and retain human confirmation for consequential operations.
Is Gemini 3.6 Flash better than Gemini 3.5 Flash?
Google’s published evaluations show higher scores in coding, machine-learning engineering, knowledge work and computer use, alongside lower output-token usage. The practical advantage depends on the workload, so teams should run representative tests before changing production traffic.
Bottom line
Gemini 3.6 Flash is a meaningful efficiency-focused upgrade rather than a cosmetic version change. It improves Google’s published results on agentic coding, machine-learning engineering, knowledge work and computer use, while cutting the standard output-token price from $9.00 to $7.50 per million tokens. Its strongest case is not simply that each token is cheaper, but that many multi-step tasks may require fewer tokens, turns and tool calls.
Developers should consider it the primary migration candidate for Gemini 3.5 Flash workloads that rely on code, tools or multimodal reasoning. Flash-Lite remains more suitable for the cheapest high-volume execution, and Gemini 3.6 Flash is not the right model for image generation, native audio output or Live API use. Benchmark gains are promising, but production decisions should still be based on task-specific accuracy, latency, safety and total cost.