[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"academy-blogs-en-1-1-all-google-gemini-3-7-flash-update-all--*":3,"academy-blog-translations-aj4nzxaokynwo83":88},{"data":4,"page":87,"perPage":87,"totalItems":87,"totalPages":87},[5],{"alt":6,"collectionId":7,"collectionName":8,"content":9,"cover_image":10,"cover_image_path":11,"cover_image_s_url":12,"created":13,"created_by":14,"expand":15,"id":81,"keywords":82,"locale":52,"published_at":83,"scheduled_at":68,"school_blog":77,"short_description":84,"status":75,"title":85,"updated":86,"updated_by":14,"slug":78,"views":80},"Google Gemini 3.7 Flash AI model cover image for developers","sclblg987654321","school_blog_translations","\u003Cp>The race in AI for coding is accelerating, and this time Google has made its next move. On August 13, 2026, Google introduced \u003Ca target=\"_blank\" rel=\"noopener\" class=\"ng-star-inserted\" href=\"https:\u002F\u002Fblog.google\u002Finnovation-and-ai\u002Fmodels-and-research\u002Fgemini-models\u002Fintroducing-gemini-3-7-flash\u002F\">Gemini 3.7 Flash\u003C\u002Fa>, a new model in the Gemini 3 family positioned specifically as a “workhorse model” for coding and AI agents. Interestingly, this launch comes just about 3 weeks after \u003Ca target=\"_blank\" rel=\"noopener\" class=\"ng-star-inserted\" href=\"https:\u002F\u002Fblog.google\u002Finnovation-and-ai\u002Fmodels-and-research\u002Fgemini-models\u002Fgemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber\u002F\">Gemini 3.6 Flash\u003C\u002Fa>.\u003C\u002Fp>\u003Cp>But Gemini 3.7 Flash is not just a version number bump. \u003Ca target=\"_blank\" rel=\"noopener\" class=\"ng-star-inserted\" href=\"https:\u002F\u002Fwww.google.com\u002F?hl=th\">Google\u003C\u002Fa> states that this new model features improvements across software engineering, web development, agentic tool use, and tasks requiring complex document or data analysis. It also comes with an introductory price set at $0.75 per 1 million input tokens and $3.75 per 1 million output tokens until the end of 2026.\u003C\u002Fp>\u003Cp>So, how much has Gemini 3.7 Flash changed from its predecessor? And if you are a developer, what should you care about? Let's take a look.\u003C\u002Fp>\u003Ch2>What is Gemini 3.7 Flash?\u003C\u002Fh2>\u003Cp>Gemini 3.7 Flash is the successor to Gemini 3.6 Flash in the Gemini 3 family. Google DeepMind notes that the model has received algorithmic improvements building upon the existing reasoning foundation of 3.6 Flash to better handle complex tasks.\u003C\u002Fp>\u003Cp>What Google clearly prioritizes with this model are coding, agentic workflows, and tasks requiring continuous multi-step execution. Simply put, instead of just using AI to ask, \"What's wrong with this code?\", the new generation of AI is being developed closer to a workflow of \"Inspect this problem → Read relevant code → Plan a fix → Use a tool → Apply fix → Verify results.\"\u003C\u002Fp>\u003Cp>On the input side, Gemini 3.7 Flash supports text, image, audio, and video. It has a context window of up to 1 million tokens and can generate text output up to 64K tokens, according to data in Google DeepMind's model card.\u003C\u002Fp>\u003Ch2>1. Improved Coding and Debugging\u003C\u002Fh2>\u003Cimg src=\"https:\u002F\u002Fpb.tumwebsme.com\u002Fapi\u002Ffiles\u002Fpbc_2997280662\u002Fr7d76ty5uh89auo\u002F4_11zon_00z2tr61vm.webp\" style=\"display: block; margin: 0px auto;\">\u003Cp>One of the areas Google emphasizes most with Gemini 3.7 Flash is software engineering. Google states that the model can handle tasks like debugging and issue resolution better, including a higher first-pass code accuracy. This means a higher likelihood that the code provided in the first response will match the intended request, based on Google's testing.\u003C\u002Fp>\u003Cp>Examples from the benchmarks revealed by Google:\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cp>\u003Cstrong>FrontierCode 1.1\u003C\u002Fstrong> Main for measuring production code quality: Gemini 3.7 Flash achieved 43.6% compared to Gemini 3.6 Flash's 34.4%.\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>\u003Cstrong>DeepSWE v1.1 \u003C\u002Fstrong>for long-horizon software engineering: 3.7 Flash achieved 65.3% while 3.6 Flash scored 48.6%.\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>\u003Cstrong>Terminal-bench 2.1\u003C\u002Fstrong> for agentic terminal coding: 3.7 Flash achieved 85.8% compared to 3.6 Flash's 78.0%.\u003C\u002Fp>\u003C\u002Fli>\u003C\u002Ful>\u003Cp>These numbers do not mean Gemini 3.7 Flash will write perfect code every time, but they reflect that on these benchmark suites, the model performs noticeably better than its predecessor.\u003C\u002Fp>\u003Cp>For developers, the interesting part is therefore not just asking the AI to \"write a function for me,\" but tackling tasks that involve more steps, such as analyzing bugs involving multiple files, reviewing a codebase, or having an agent help manage issues continuously across multiple steps.\u003C\u002Fp>\u003Ch2>2. Web Development is Another Upgraded Area\u003C\u002Fh2>\u003Cp>Frontend and web developers will likely be particularly interested in this part. Google states that Gemini 3.7 Flash can generate more functional web applications with fewer prompts. It also improves design adherence, meaning its ability to generate web pages that closely match a provided design reference. References provided to the model can be screenshots, images, or a full design system.\u003C\u002Fp>\u003Cp>Another use case Google directly mentions is having the model review an existing codebase and compare it against a mockup to check how closely the implementation matches the design. On \u003Ca target=\"_blank\" rel=\"noopener\" class=\"ng-star-inserted\" href=\"https:\u002F\u002Farena.ai\u002Fleaderboard\u002Fcode\u002Fwebdev\">WebDev Arena\u003C\u002Fa>, the Gemini 3.7 Flash model achieved an Elo score of 1588, up from Gemini 3.6 Flash's 1538, according to the benchmarks published by Google. For frontend developers, this could make the \"Design → Prompt → Generate → Compare → Fix\" workflow more compelling, especially for landing pages, dashboards, or UIs with clear design references.\u003C\u002Fp>\u003Cblockquote>\u003Cp>\u003Cem>Of course, AI-generated code should still be reviewed for responsiveness, accessibility, performance, security, and business logic by a developer before being used in production.\u003C\u002Fem>\u003C\u002Fp>\u003C\u002Fblockquote>\u003Ch2>3. Better Multi-step AI Agent Workflows\u003C\u002Fh2>\u003Cp>Another term Google frequently mentions in this launch is \"Agentic Workflows.\" An AI agent differs from general question-and-answer chats in that an agent can execute a sequence of multiple steps, use tools, and decide what to do next based on previous outcomes.\u003C\u002Fp>\u003Cp>Google notes that Gemini 3.7 Flash improves multi-step planning and tool calling, as well as handling roadblocks during execution better. It follows instructions more accurately and can ask questions to clarify intent when necessary. In benchmarks like \u003Ca target=\"_blank\" rel=\"noopener\" class=\"ng-star-inserted\" href=\"https:\u002F\u002Fzapier.com\u002Fblog\u002Fintroducing-automationbench\u002F\">AutomationBench\u003C\u002Fa>, which tests automation in enterprise workflows, Gemini 3.7 Flash scored 30.4% compared to Gemini 3.6 Flash's 17.0%.\u003C\u002Fp>\u003Cp>Furthermore, Google is making Gemini 3.7 Flash the default model for the Antigravity Agent in Gemini Managed Agents and the Google Antigravity SDK. This clearly reflects Gemini's direction: Google no longer views AI purely as a chatbot waiting for prompts, but is developing models to be part of workflows that can seamlessly operate with tools and other systems.\u003C\u002Fp>\u003Ch2>4. Selectable \"Thinking\" Levels Based on the Task\u003C\u002Fh2>\u003Cp>Another interesting capability of Gemini 3.7 Flash is that developers can choose the model's \"Thinking\" level across 3 options:\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cp>\u003Cstrong>Low\u003C\u002Fstrong>: For tasks requiring low latency, such as real-time chat, incident response, or rapid data analysis.\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>\u003Cstrong>Medium\u003C\u002Fstrong>: The default setting, suitable for general tasks, including coding and moderately complex agents.\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>\u003Cstrong>High\u003C\u002Fstrong>: For reasoning, mathematics, or more difficult coding and agentic tasks, which may consume more tokens and incur higher costs.\u003C\u002Fp>\u003C\u002Fli>\u003C\u002Ful>\u003Cp>This is important because developers don't need the model to apply maximum reasoning to every request. If it's a simple task like formatting data or generating short text, you can select \"Low\" to prioritize speed. But if you are having an agent analyze a problem in a large-scale system, you can increase the thinking level, trading latency and cost for higher reasoning capabilities.\u003C\u002Fp>\u003Cblockquote>\u003Cp>\u003Cem>In other words, we have more choices in deciding which tasks need to be \"fast\" and which tasks need to \"think thoroughly.\"\u003C\u002Fem>\u003C\u002Fp>\u003C\u002Fblockquote>\u003Ch2>5. Reads Data Across Text, Image, Audio, and Video\u003C\u002Fh2>\u003Cp>Gemini 3.7 Flash remains a multimodal model, supporting inputs of text, image, audio, and video, with a context window of up to 1 million tokens. This means its use cases are not limited to just coding.\u003C\u002Fp>\u003Cp>Developers might use the model to analyze UI screenshots, read massive amounts of documentation, analyze multiple file types combined, or build agents that need to process information in various formats. In Google's evaluation results, the model was also assessed on long context, long video understanding, and synthesizing data from graphs or complex documents.\u003C\u002Fp>\u003Ch2>6. How Much Does Gemini 3.7 Flash Cost?\u003C\u002Fh2>\u003Cp>Another point just as interesting as its capabilities is the price. Google has set the introductory pricing for Gemini 3.7 Flash at:\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cp>\u003Cstrong>Input\u003C\u002Fstrong>: $0.75 per 1 million tokens\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>\u003Cstrong>Output\u003C\u002Fstrong>: $3.75 per 1 million tokens\u003C\u002Fp>\u003C\u002Fli>\u003C\u002Ful>\u003Cp>This is an introductory pricing rate and is valid only until December 31, 2026.\u003C\u002Fp>\u003Cp>Starting January 1, 2027, Google indicates the pricing will change to:\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cp>\u003Cstrong>Input\u003C\u002Fstrong>: $1.50 per 1 million tokens\u003C\u002Fp>\u003C\u002Fli>\u003Cli>\u003Cp>\u003Cstrong>Output\u003C\u002Fstrong>: $7.50 per 1 million tokens\u003C\u002Fp>\u003C\u002Fli>\u003C\u002Ful>\u003Cblockquote>\u003Cp>\u003Cem>Therefore, if you are evaluating costs for a long-term production system, you should calculate based on the post-introductory pricing as well, not just the launch price.\u003C\u002Fem>\u003C\u002Fp>\u003C\u002Fblockquote>\u003Ch2>Where Can Developers Use Gemini 3.7 Flash?\u003C\u002Fh2>\u003Cp>Developers can start using Gemini 3.7 Flash via the Gemini API and Google AI Studio, as well as in Android Studio and agent-first workflows via Google Antigravity. The Model ID for the API is \u003Ccode>gemini-3.7-flash\u003C\u002Fcode>.\u003C\u002Fp>\u003Cp>For enterprises, it can be accessed through the Gemini Enterprise Agent Platform and the Gemini Enterprise App. For general users, Google is bringing this model to Gemini Spark for Google AI Pro and Ultra subscribers in supported countries.\u003C\u002Fp>\u003Ch2>Things to Know Before Using Gemini 3.7 Flash\u003C\u002Fh2>\u003Cp>Even though many benchmark scores for Gemini 3.7 Flash have improved, it does not mean the AI can entirely replace developers without oversight.\u003C\u002Fp>\u003Cp>Google DeepMind directly notes in the \u003Ca target=\"_blank\" rel=\"noopener\" class=\"ng-star-inserted\" href=\"https:\u002F\u002Fdeepmind.google\u002Fmodels\u002Fmodel-cards\u002Fgemini-3-7-flash\">model card\u003C\u002Fa> that Gemini 3.7 Flash still carries the general limitations of foundation models, including hallucinations or fabricating incorrect information, and there may be instances of latency or timeouts.\u003C\u002Fp>\u003Cp>Thus, if using it in production especially for generating code, modifying systems, automating tasks, or letting agents use tools there should be a process to verify the results. Proper testing and permissions must be in place before letting AI execute actions on critical systems. Benchmarks themselves should be viewed merely as one set of metrics, because real-world results still depend on the codebase, prompts, tools, context, and the specific nature of each team's work.\u003C\u002Fp>\u003Ch2>How Compelling is Gemini 3.7 Flash for Developers?\u003C\u002Fh2>\u003Cp>Looking at what Google upgraded in this release, the standout feature of Gemini 3.7 Flash isn't just one specific capability. Rather, it's the attempt to combine Coding + Reasoning + Multimodal + Tool Use + Agentic Workflows into a \"Flash\" model designed for wide-scale deployment.\u003C\u002Fp>\u003Cp>Those working in web development or experimenting with AI agents might see the benefits quite clearly, as Google has directly improved code generation, designing from references, and multi-step agent workflows.\u003C\u002Fp>\u003Cp>What is worth watching next, then, is perhaps not just \"how many percent smarter\" the next version of Gemini will be, but how far developers can let the AI run continuously before a human needs to intervene. As AI moves beyond merely helping write code to being able to read designs, analyze codebases, utilize tools, and execute multiple steps simultaneously, the developer workflow might change quite a bit.\u003C\u002Fp>\u003Cdiv data-type=\"horizontalRule\">\u003Chr>\u003C\u002Fdiv>\u003Ch2>FAQ: Frequently Asked Questions About Gemini 3.7 Flash\u003C\u002Fh2>\u003Ch3>When was Gemini 3.7 Flash released?\u003C\u002Fh3>\u003Cp>Google officially launched Gemini 3.7 Flash on August 13, 2026, releasing it just about 3 weeks after \u003Ca target=\"_blank\" rel=\"noopener\" class=\"ng-star-inserted\" href=\"https:\u002F\u002Fblog.google\u002Finnovation-and-ai\u002Fmodels-and-research\u002Fgemini-models\u002Fgemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber\u002F\">Gemini 3.6 Flash\u003C\u002Fa>.\u003C\u002Fp>\u003Ch3>What tasks is Gemini 3.7 Flash best suited for?\u003C\u002Fh3>\u003Cp>Google recommends the model for tasks like code generation, spatial and multimodal reasoning, multi-step agentic workflows, and tasks that specifically require design adherence.\u003C\u002Fp>\u003Ch3>What is the context window for Gemini 3.7 Flash?\u003C\u002Fh3>\u003Cp>The model supports a context window of up to 1 million tokens and a maximum text output of 64K tokens. It can accept inputs in text, image, audio, and video formats.\u003C\u002Fp>\u003Ch3>How much does Gemini 3.7 Flash cost?\u003C\u002Fh3>\u003Cp>The introductory pricing is $0.75 per 1 million input tokens and $3.75 per 1 million output tokens until December 31, 2026. This will adjust to $1.50 and $7.50 respectively starting January 1, 2027.\u003C\u002Fp>\u003Ch3>Does Gemini 3.7 Flash always write correct code?\u003C\u002Fh3>\u003Cp>No. Even though multiple coding benchmark results have improved from Gemini 3.6 Flash, Google DeepMind still notes that the model can experience hallucinations. Therefore, AI-generated code and answers should always be reviewed and tested before actual implementation.\u003C\u002Fp>\u003Cdiv data-type=\"horizontalRule\">\u003Chr>\u003C\u002Fdiv>\u003Ch2>Conclusion\u003C\u002Fh2>\u003Cp>Gemini 3.7 Flash is an upgrade to Google's Flash Series that clearly prioritizes coding and AI agents. This includes bug fixing, long-horizon software engineering, web development, working from design references, and multi-step tool usage.\u003C\u002Fp>\u003Cp>For developers, the most compelling aspect might not just be the higher benchmark scores, but the direction in which AI is shifting: from a \"coding question-and-answer assistant\" to a \"workflow assistant capable of executing multiple steps.\" Yet, AI still has limitations and can provide incorrect answers. Reviewing, testing, and controlling agent permissions remain crucial parts of deploying these models in reality.\u003C\u002Fp>\u003Cp>What about you? Between a fast and affordable AI, and one with deeper reasoning that takes more time and tokens, which would you choose for your daily workflow? Let us know in the comments below! 👇 And if you don't want to miss out on fresh tech updates like this, make sure to follow Superdev Academy\u003C\u002Fp>","1c8vtfgndre_mxff2aoir4_edj4absn2o.webp","https:\u002F\u002Ftwsme-r2.tumwebsme.com\u002Fsclblg987654321\u002Fhpabh9cxcsg3xew\u002F1c8vtfgndre_mxff2aoir4_edj4absn2o.webp","https:\u002F\u002Ftwsme-r2.tumwebsme.com\u002Fsclblg987654321\u002Fhpabh9cxcsg3xew\u002Fs\u002F1c8vtfgndre_mxff2aoir4_edj4absn2o.webp","2026-08-18 09:14:34.640Z","423vhnv3ckczcyn",{"keywords":16,"locale":46,"school_blog":56},[17,23,27,33,38,42],{"collectionId":18,"collectionName":19,"created":20,"created_by":14,"id":21,"name":22,"updated":20,"updated_by":14},"sclkey987654321","school_keywords","2026-08-18 09:06:35.607Z","dj4jb1aw6nqunp3","Gemini 3.7 Flash",{"collectionId":18,"collectionName":19,"created":24,"created_by":14,"id":25,"name":26,"updated":24,"updated_by":14},"2026-08-18 09:14:06.480Z","pog2zzt2qxj9zfi","Google DeepMind",{"collectionId":18,"collectionName":19,"created":28,"created_by":29,"id":30,"name":31,"updated":32,"updated_by":29},"2026-05-21 16:32:55.489Z","76qprkevbgfdps8","rdv13qjgom4ml7p","AI Coding","2026-06-07 06:49:18.676Z",{"collectionId":18,"collectionName":19,"created":34,"created_by":29,"id":35,"name":36,"updated":37,"updated_by":29},"2026-03-04 08:44:26.139Z","dlm8aajwkiz9tae","AI Agent","2026-06-07 06:46:33.227Z",{"collectionId":18,"collectionName":19,"created":39,"created_by":29,"id":40,"name":41,"updated":39,"updated_by":29},"2026-06-11 16:14:34.250Z","01ajl5eq1joxocg","LLM",{"collectionId":18,"collectionName":19,"created":43,"created_by":14,"id":44,"name":45,"updated":43,"updated_by":14},"2026-08-18 09:14:30.463Z","isr7idmz3xt5xxw","Web Development AI",{"code":47,"collectionId":48,"collectionName":49,"created":50,"flag":51,"id":52,"is_default":53,"label":54,"updated":55},"en","pbc_1989393366","locales","2026-01-22 11:00:02.726Z","twemoji:flag-united-states","qv9c1llfov2d88z",false,"English","2026-04-10 15:42:46.825Z",{"category":57,"collectionId":58,"collectionName":59,"created":60,"expand":61,"id":77,"slug":78,"updated":79,"views":80},"jwhmeg2uxwz0bkp","pbc_2105096300","school_blogs","2026-08-18 09:07:15.631Z",{"category":62},{"blogIds":63,"collectionId":64,"collectionName":65,"created":66,"created_by":29,"id":57,"image":67,"image_alt":68,"image_path":69,"image_s_url":70,"label":71,"name":72,"priority":73,"publish_at":74,"scheduled_at":68,"status":75,"updated":76,"updated_by":29},[],"sclcatblg987654321","school_category_blogs","2026-03-04 08:31:40.635Z","1qt33a4n2u3_lr23qy62xh_5wkvdn18n1.webp","","https:\u002F\u002Ftwsme-r2.tumwebsme.com\u002Fsclcatblg987654321\u002Fjwhmeg2uxwz0bkp\u002F1qt33a4n2u3_lr23qy62xh_5wkvdn18n1.webp","https:\u002F\u002Ftwsme-r2.tumwebsme.com\u002Fsclcatblg987654321\u002Fjwhmeg2uxwz0bkp\u002Fs\u002F1qt33a4n2u3_lr23qy62xh_5wkvdn18n1.webp",{"en":72,"th":72},"Tools & Resources",0,"2025-01-31 09:55:27.001Z","published","2026-08-18 14:49:09.952Z","aj4nzxaokynwo83","google-gemini-3-7-flash-update","2026-08-19 13:14:04.026Z",120,"hpabh9cxcsg3xew",[21,25,30,35,40,44],"2026-08-19 04:23:56.406Z","Discover Gemini 3.7 Flash by Google. Upgraded AI for coding, web development, and agentic workflows with better reasoning and lower costs.","Google Introduces Gemini 3.7 Flash The Ultimate AI Model for Coding and Agents","2026-08-19 04:24:03.237Z",1,{"th":78,"en":78}]