Gemini 3.5: frontier intelligence with action

https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/

68•meetpateltech•38m ago

Comments

alexdns•28m ago

Its Gemini 3.5 Flash

nerdalytics•7m ago

Yeah, Google chose a misleading title for the blog post.

asar•23m ago

$1.5/m input tokens $9/m output tokens

6x the price of 3.1 flash lite

himata4113•21m ago

I don't think input/output pricing matters, 90% of the cost is cache. $0.15 is pretty good, but still very expensive.

minimaxir•20m ago

10% of input pricing is standard especially compared to competition.

himata4113•12m ago

yah, which means that the input cost is the only value that should be paid attention to at the end + the cache discount (x10). If google would start offering x20 discount it would make it twice as cheap while input and output stayed the same.

wolttam•10m ago

It depends on the use-case. yes, 90% of cost is cache in agentic coding scenarios (actually 95% in my experience). But not when the model reasons for 200k+ tokens before answering a complex problem.

__jl__•8m ago

In our experience, caching is not very reliable with google. We always get random cache misses that don't happen with other providers. We find OpenAI, Anthropic and Fireworks (which we use a lot) all have higher cache hit rates. So it's not only about the costs of cached token but also what kind of cached hit rate you get.

John7878781•20m ago

Sounds reasonable given the difference in model performance.

Edit: Never mind.

stri8ed•17m ago

Output cost is 3x from Gemini 3 flash.

iwhalen•18m ago

I wonder why they didn't discuss price in the post?

Compare to the GPT-5.5 announcement: https://openai.com/index/introducing-gpt-5-5/

himata4113•22m ago

Engineers at google have publically stated that the models are too big and are far from their potencial. Glad they're being proven right with every release.

They continue to focus on smaller models while openai and anthropic are increasing compute requirements for their SOTA models.

stri8ed•18m ago

Given the cost increase associated with this model, and previous model releases, I think the size is trending upwards, not down.

himata4113•13m ago

The speed says otherwise. I think they're increasing costs since they want to start seeing ROI.

maipen•16m ago

Don’t let that fool yourself. Google will have SOTA models as big as or even bigger than their competitors.

They are just refining their current models while they finish training the next generation.

They will all come out at about the same time. Anthropic, OpenAi, Google, xAI

ACCount37•9m ago

Anthropic has been sitting on Mythos for a while now. I guess they don't feel pressured to fuck it ship it until anyone else gets a 10T to work.

howdareme•10m ago

Google’s pro models are almost certainly bigger than Openai’s lol

golfer•21m ago

Here's the benchmark scoreboard they published:

https://storage.googleapis.com/gweb-uniblog-publish-prod/ori...

mixtureoftakes•16m ago

benchmarks look REALLY good, the price hike is big but it also beats sonnet 4.6 in every discipline?

postalcoder•13m ago

i have a hard time believing Google employees use their own harness. Antigravity suffers from a well known bug where it totally breaks when git extensions.worktreeConfig is true. Like it's completely broken. This has been an issue for at least two months.

Google employees, perhaps nudge folks in the org to fix this bug?

SXX•9m ago

  > Create animated SVG of a frog on a boat rowing through jungle river. Single page self contained HTML page with SVG

3.5 Flash: Thinking Medium - 7516 tokens

https://gistpreview.github.io/?5c9858fd2057e678b55d563d9bff0...

3.5 Flash: Thinking High - 7280 tokens

https://gistpreview.github.io/?1cab3d70064349d08cf5952cdc165...

3.1 Pro - 28,258 tokens

https://gistpreview.github.io/?6bf3da2f80487608b9525bce53018...

Though 3.1 took 3 minutes of thinking to generate, but it only one that got animated movement.

abi•7m ago

Your links are broken FYI.

John7878781•7m ago

They work for me.

cesarvarela•8m ago

Add Flash to the title, please.

benbencodes•7m ago

Pricing is now live on ai.google.dev/pricing:

Gemini 3.5 Flash: $0.75 input / $4.50 output per 1M tokens, 1M context window. The output price explicitly "includes thinking tokens" — which is why it's higher than a typical flash-class model.

For comparison within the Gemini lineup: - Gemini 2.5 Flash: $0.30 / $2.50 - Gemini 3.1 Flash-Lite: $0.25 / $1.50 - Gemini 3.1 Pro Preview: $2.00 / $12.00

So 3.5 Flash is ~2.5x more expensive input vs 2.5 Flash. The pricing and "including thinking tokens" framing position it as a reasoning-capable flash model rather than just a pure speed optimization.

aliljet•7m ago

Is there a good benchmark tracking hallucinations? The models are all incredibly good now, even the open ones, and my hope is that the rate of hallucinations is something that's falling off in concert with larger and larger context lengths.

bakugo•5m ago

Triple the price of the last Flash model ($3 -> $9 per 1M output). Quickly approaching Sonnet prices.

Feels like the AI pricing noose is tightening sooner rather than later.

Intentional Amnesia: Why I Wipe My AI Agents' Memories Every 15 Minutes

$400 Audemars Piguet Swatch

O Americano, Outra Vez (By Richard P. Feynman)

Tonic is joining the gRPC project

Meta layoffs starting this week stress harsh reality inside Zuckerberg company

Ask HN: Have I Become a Luddite?

Understanding the Go Runtime: The Select Statement

PostgreSQL 19 Beta: The Four Features You'll Feel

Ask HN: Why invoice matching still pain in the ass in 2026?

Show HN: I reviewed 271 event tech tools and turned the data into a visual story

Microsoft's Durabletask Package on PyPI Compromised. Mini Shai Hulud

Graduates are booing pep talks on AI at college commencements

Antigravity 2.0

What We Talk About When We Talk About the Weather

U.S. Authorities Move to Resolve Gautam Adani's Legal Woes

Using WebAssembly SIMD to Fingerprint CPU Models from the Browser

Show HN: Bench-Bets – A World Cup Prediction App Built with Nuxt and SQLite

There Shall Be Cathedrals

Introducing Kiro Web: Build, delegate, and steer right from the browser

Show HN: AgentShield – Stop AI agents from spending money unsupervised

Gemini Omni

Wild Young People

Deciphering the Hashihara Castle Town Map

On Blind, Anxious Tech Workers Get the Lowdown on Layoffs

The Iranian Government Filtering Cloud Is Trying to Get Into Mozilla

Google Changes Its Search Box for the First Time in 25 Years

Gemini CLI will stop working from June 18, 2026

Cat Organ

Gemini Spark

Google Antigravity 2.0