Google's fast multimodal reasoning model with million-token context, complex coding, and agent execution for software development, cross-media analysis, and multi-step automation.
| Item | Specification |
|---|---|
| Input | Text, image, video, audio, PDF |
| Output | Text |
| Context | 1,048,576 tokens |
| Max Output | 65,536 tokens |
| Vision | ✓ Supported |
| Function Calling | ✓ Supported |
Gemini 3.6 Flash is a stable multimodal language model released by Google on July 21, 2026 under the model code gemini-3.6-flash. It targets fast reasoning and agent loops for real-world tasks, accepts text, image, video, audio, and PDF inputs, produces text, and provides a 1,048,576-token input window with up to 65,536 output tokens.
Compared with Gemini 3.5 Flash, Google focuses this release on more concise output, code generation, agent planning, and spatial reasoning. It supports thinking, function calling, code execution, File Search, Google Search and Maps grounding, URL Context, structured output, and caching. Computer Use is supported, while image generation, audio generation, and the Live API are not supported by this model.
| Feature | Benefit |
|---|---|
| Native Multimodal Input | A single model processes text, images, video, audio, and PDFs, reducing conversion and orchestration overhead. |
| 1,048,576-Token Context | Preserves relationships across large projects and long cross-media source collections. |
| Configurable Thinking | Applications can align reasoning depth with task complexity and interaction requirements. |
| Built-In Tool Ecosystem | Search, Maps, file retrieval, URL Context, and code execution connect answers to external evidence and real results. |
| Structured Output and Function Calling | Supports integration with agent orchestrators, business APIs, and automated validators. |
| Stable Model Endpoint | A fixed model code supports reproducible tests and version management without relying on a moving latest alias. |
Start with the task, a source index, and evidence priorities, then group material by document section, code module, or media segment. Remove duplicates and request filenames, page numbers, or timestamps so conclusions can be checked.
Provide the target viewport, full-page references, and explicit typography, spacing, and component requirements. Compare rendered screenshots and interactions before revising individual issues, rather than judging success only by whether the code runs.
Test normal tool responses, empty results, and tool failures. Check argument validity, use of returned evidence, and final task completion separately, validating important fields in application code.
Record both the requested model identifier and returned version information rather than relying on a display name. Google still offers 3.6 Flash for production, while its 3.7 Flash requests automatically route to 3.8 Flash; confirm LinkModel's mapping separately in its documentation.
Open the corresponding LinkModel page and configure the published model identifier, authentication method, and request format. Validate a short text request first, then configure multimodal input, tools, and reasoning within the documented support scope.
Test representative text, multimodal, long-context, and multi-turn tool workflows. Record response versions, duration, stop reasons, and task success, checking supported parameters without assuming Google-native fields pass through unchanged.
Our pricing is based on image and text token usage. The final cost depends on the tokens consumed.
| Token Type | LinkAI Price | Official Price |
|---|---|---|
| Input | $0.75 / 1M tokens | $1.5 / 1M tokens |
| Cached input | $0.075 / 1M tokens | $0.15 / 1M tokens |
| Output | $3.75 / 1M tokens | $7.5 / 1M tokens |
| Reasoning output | $3.75 / 1M tokens | $7.5 / 1M tokens |