DoKoB

@sundarpichai @antigravity The pricing should be around $3.5/M tokens, not $9/M tokens. This model is on-par with Kimi K2.6 which is at around $3.5/M tokens.

Just off stage at #GoogleIO, some highlights from this morning 🧵 Gemini 3.5 Flash is available today for everyone in @antigravity and across our products and APIs. Compared to 3.1 Pro, 3.5 Flash is better across almost all benchmarks with huge progress in coding. It’s also comparable to the best models but very fast (4x faster tokens/ second than other frontier models). And when looking at the intelligence versus output speed, it’s in a league of its own in the top right quadrant.

@OfficialLoganK Nice progress on the pricing for the output token as well.

Today we are starting to roll out the biggest upgrade to the Google Search box in over 25 years — now completely reimagined with AI, along with Gemini 3.5 Flash as the new default model for AI mode users globally!

@OfficialLoganK The price is the real deal-killer here. 3x from the previous model. Flash and Flash-lite were suppose to be the affordable options.

@JeffDean $9/M tokens - 3x the price from the previous model too.

1/ Today at #GoogleIO, we’re releasing Gemini 3.5, our latest family of models combining frontier intelligence with action. We’re starting by releasing 3.5 Flash, which is built to help you execute complex, long-horizon agentic workflows. Gemini 3.5 Flash is our strongest model for coding and agent yet.It outscores 3.1 Pro on agentic and coding benchmarks like Terminal-Bench and MCP Atlas, while running 4x faster than other frontier models. Used in Google Antigravity, 3.5 Flash is even further optimized to be up to 12x faster. It’s a powerful engine to deploy sub-agents that collaborate, run high-frequency iterative loops, and solve real-world problems at scale. Some highlights we’re excited about 🔽

@OfficialLoganK @GoogleDeepMind Make it more affordable tho. $9/M tokens is really expensive compared to the previous flash model.

Welcome to Gemini 3.5 Flash, our most powerful model to date. It pushes the frontier of intelligence, speed, and cost putting 3.5 Flash in a class of its own. We spent the last 6 months making sure Flash is great for real world use cases. It's available everywhere now!

@cerebras @ArtificialAnlys @Kimi_Moonshot Does it scale tho? You guys need so many DCs to run and serve 3-5 models.

Cerebras is now running Kimi K2.6 – a trillion parameter model – in enterprise trials. At ~1,000 tokens/s, this is the fastest frontier model performance ever measured by Artificial Analysis @ArtificialAnlys.

@EdenKollcinaku @GoogleDeepMind @GoogleAI Wait until they nerf the model before releasing into GA.

i tried Gemini 3.5 Flash inside Antigravity. trust me: it makes claude code/codex look nerfed. kudos to @GoogleDeepMind @GoogleAI for delivering such a great fast model imagine how 3.5 Pro will feel like...

@harukaawake They are going to multiply, while the natives are going to get eradicated by them. Soon their rules will apply to streets and culture as well. Watch what happened to UK and other EU countries. You have to place strict rules to save your culture and country from these mfs.

🇯🇵 Many Japanese people are taking note that while most schools serve pork, some schools seem to be avoiding it to not offend Muslims. In Arakawa Ward, for example, one school serves chicken 36 times and pork only 10 times.

@DNAutics Welp, my gay-dads are fighting once again.

CEO: you dont understand open source. me: ....?

@jarredsumner Blog post with result findings or not happened.

99.8% of bun’s pre-existing test suite passes on Linux x64 glibc in the rust rewrite

@sudox7 Ugly C? This is the only for loop I know and use (even when writing JS/TS).

your "clean" C++ is slower than this "ugly" C

@AaryanPrateekX India is a joke.. Super power by 2040-2050 my ass. The whole country is corrupted and hollow from the inside.

Laude ka LAW Laude ka Country The most corrupt institution in this country is the judiciary and the courts. The first reform this country needs is to nuke its courts.

@hotahotasa45082 Stop accepting so many muslims from these "Muslim majority" countries.

←私が望んでる日本 　　　　　　　　　日本政府が望んでる日本→

AI coding models will spawn a generational surge of knowledge-gaps and tech-debt over the next decade and the aftermath will be... messy, massive, and unavoidable.

@URANUS0509 @bcfe70bord Did by muslims for sure. They don't like other religion and want their to be everywhere.

@bcfe70bord 外人が全部やったって証拠でもあるのか? 日本人だって、落書きやら食べ物粗末に扱うのに？ そんな事を軽く言うから反発も生まれるわ。 俺達の物言いが生んだ結果かもしれないよ？

やっぱり絶対におかしい‼️ 昨年からの火事がこれまでよりも多すぎる‼️ 新潟県中央区の300年前からある由緒ある愛宕神社が、7日の深夜に突然火事に。。。 私は放火としか思えない。 最近、養豚場や山の火事、墓跡荒らしなど日本の大事な物が壊されすぎ。 制限無く移民や外国人観光客を連れてくるからでは？

@Galaxia0xFF If you can't handle these kinds of cases when writing in C, you shouldn't be writing C.

how three languages handle a missing value C will kill you silently C++ will kill you politely Rust won't let you write the code at all production bugs disagree on which is best

@jarredsumner How about @AnthropicAI creates a model that's specialized on Ziglang? That can help you guys a ton (and others too).

why: I am so tired of worrying about & spending lots of time fixing memory leaks and crashes and stability issues. it would be so nice if the language provided more powerful tools for preventing these things.

@jarredsumner @Vishal_anton16 I guess if the rust version works same as the zig version but with comparable performance and low chances crashes and memory leak issues, bun will switch to it.

@Vishal_anton16 No. It’s sometimes I wish you could have destructors and constructors. A borrow checker sure sounds nice. Maybe we’d have fewer github issues if these things existed in the language.

Zig has been great for Bun. This branch is a fun experiment, and not a decision - here’s what I wrote on HN:

@saltyAom What's a plugin in elysia?

Elysia 2α Apply 100k plugin with 1 route and a "scoped" lifecycle event 44GB → 35MB • memory usage 196,197ms → 14.75ms • time taken

@saltyAom @jarredsumner Summon robobun to optimize this asap.

Hmm... Bun.serve.routes use more memory than I expected There seems to be some exponential delay before starting the server, based on the total routes 1k: 9ms 5k: 79ms 10k: 257ms 20k: 694ms 40k: 3117ms 100k: 22,424ms 100k is impractical, but it proves something is going on