We’re introducing Gemini 3.8 Flash ⚡️ built to tackle complex agentic and multi-step tasks with even greater diligence. Our most intelligent workhorse model yet delivers significant improvements in reasoning, evolving to an AI partner that doesn’t just write code, but can also navigate complex projects. While solving ambiguous and... high-friction tasks is immensely helpful, it can also be expensive. Fortunately, 3.8 Flash features the usual effort controls, ensuring that the amount of thinking required to accomplish the task at hand is proportional to the tokens spent. Watch Gemini 3.8 Flash combine native video understanding with advanced coding to autonomously build this 3D game, play it to find errors, and execute code changes in a seamless agentic loop in Google Antigravity.show more

bro you had infinity tokens and this is what you came up with?

— Available to Google AI Pro and Ultra subscribers across the @GeminiApp, AI Mode in Google Search, and Google Sheets — Build in the Gemini API via @GoogleAIStudio and @AndroidStudio, explore agent-first workflows in @antigravity, and generate UIs with 3.8 Flash in @stitchbygoogle — Gemini Enterprise users by picking it from the drop down model menu or in Gemini Enterprise Agent Platform.

This is the game I built in three days using Unity and Blender MCP via Antigravity, powered by Gemini 3.7 Flash. I can't wait to further develop it using 3.8 Flash.

Using Gemini 3.8 flash model in antigravity ide

@no_stp_on_snek Here are some of mine

very great models! it’s very cheap and beats Claude Opus Feels crazy But we’re waiting for PRO model!

Getting there! 💪

I've built my semantic video editor with Codex. Maybe I should build a Gemini version 🤔

Have it bring back Google Reader and then I'll be impressed.

I’m curious how it compares to Fable 5.1. Which model is actually more capable, and how big is the gap?

@GoogleAI Which metric improved most in Gemini 3.8 Flash: task success, recovery from tool errors, or fewer unnecessary steps?

More games? sooo exciting....

A demo loop is not a partner. Gemini 3.8 Flash is a language model with video in and code out. You showed it building a toy game, playing it, and patching. That is a scripted agent cycle on a cheap SKU. Calling it an “AI partner” is the same lie as every other lab: next-token machinery wearing a job title. Effort controls do not make it diligent. They make the bill proportional to how long you let it ramble. Diligence would be holding the spec, checking what it broke, and not spending another stack of tokens to invent the next error. I expected better from Google than a six-week Flash and a cinematic. The people doing real work are not waiting on your workhorse graphic. They already left for stacks that hold. Gemini is not behind because it lacks a video. It is behind because it is still an LLM you have to babysit, sold as intelligence. Persistent machine intelligence stays under human direction. Your loop is still a generator with a game clip.

Gemini 3.8 Flash free is live on us ⚡️

This is neat. Is the price temporary? Is it going to 10x after January?

Flash 4.0 soon? Common @GoogleAI are you sleeping on Gemini pro?

long loop to keep alive! 3.8 Flash is available on Atomic Bot from today, so you can put Hermes, OpenClaw or Codex on it and let a run like that go for hours.

Within the next 2 years everyone will be able to make their own high quality, studio level games with little effort. And I cant wait for that future

This is a game they built after 24 hours of prompting it to fix issues. I've seen tests from this model, and it can't do nearly as good as this in real world scenarios and outputs.

By Andraste's wet socks!👀

This is an exciting release! Alloy Alpha has features powered by Gemini Flash since the very beginning.

Diligence on multi step tasks is the right thing to optimize since most agent failures are patience rather than intelligence. How is it at backtracking?

Reasoning gains and better persistence on multi-step tasks? That's the combo that actually matters for real agentic use cases, not just leaderboard flexing.

Well darn just burned my weekly quota lastnight! How about a friendly neighborhood reset!

The 3D demo is impressive, but can Gemini 3.8 carry a messy project from brief to stable release without constant babysitting?

Let's go!

发布页写的是 Agent 与编码提升，但模型卡还有一行更值得非英语市场看：多语种自动安全评测相对 3.7 Flash 退化 5.4 个百分点（越低越好）。工业部署不能直接继承英文安全结论，中文指令、日志/OCR 文本和提示注入都应单独做红队。

Gemini 3.8 Flashの要点をまとめると、複雑な作業も丁寧にこなす・コードだけでなくプロジェクト全体を進められる・タスクの難易度で思考量を自動調整、の3つがすごいポイントだよ！ Quick summary: Gemini 3.8 Flash handles complex agentic tasks with adaptive thinking effort. Impressive!

@daradoescode we need your benchmark rn owo

What the fuck lol

effort controls are the real sleeper feature here. people see "complex agentic" and think it's gonna be expensive, but proportional thinking is exactly what makes this actually usable in prod. also that demo of it catching its own game bugs is wild.

I think I will still prefer 20 years of passion and dedication converted into game like Path of Exile

nice

For agentic coding loops, the useful baseline may be a trace of checkpoints: tests, recovery attempts, and regressions after changes. What would you log first to compare effort settings fairly?