Ivars ✨

Hype is free. Tokens are not. I test before I trust.

[Ivars ✨ @IvarsPlisko]

I needed a local AI to turn scanned invoices into clean JSON — for real work, on one 20GB GPU. No cloud, no data leaving the box.

The internet screamed: "use Gemma (@googlegemma) , it's fast and powerful."

So I did. Then I ran it on 88 real docs 🧵

[Ivars ✨ @IvarsPlisko]

@nikitabier The product really turned out great. Thank you 💪🏼

[Ivars ✨ @IvarsPlisko]

@HarshithLucky3 @Waguri_Kaoruko8 Don't pay any attention to him. He's only writing this to get views and subscribers. He's also promoting some other profile.

[Ivars ✨ @IvarsPlisko]

@kimmonismus To be honest, yes. Opus has become just awful.

[Ivars ✨ @IvarsPlisko]

@Alibaba_Qwen 3.8-Max just dropped. 2.4T parameters, open weights next week, multi-day autonomous coding runs…
Is this still “catching up” or are we watching someone quietly take the high ground one precise release at a time?

[Qwen @Alibaba_Qwen]

Meet Qwen3.8-Max: A New Bar for Coding and Cowork.

[Ivars ✨ @IvarsPlisko]

Frontier performance is nice.
Frontier performance at $0.28 per million tokens is a different game.
@deepseek_ai V4 Flash just made “good enough” look suspiciously close to “best available” for a lot of real work.

[Mia @MiaAI_lab]

DeepSeek v4 Flash pricing is wild 😮

For the same task, it's:

28x cheaper than Kimi K3
62x cheaper than GPT-5.6 Sol
78x cheaper than Opus 5
105x cheaper than Fable 5

Source: https://t.co/2FRDmS44IF

[Ivars ✨ @IvarsPlisko]

Bold claim.
Haven’t tested it yet.
Who’s already tried Qwen3.8-Max — is it actually better or just more nice words and charts?

[Qwen @Alibaba_Qwen]

Meet Qwen3.8-Max: A New Bar for Coding and Cowork.

[Ivars ✨ @IvarsPlisko]

Almost 800k people were ready to download a separate Google AI Studio app.
But Google, as always, decided to do it their way: “No need for another app, let everything just appear inside Gemini.”
They overthink everything — and that’s why they keep having release problems 😅

[Google AI Studio @GoogleAIStudio]

thank you to the ~800,000 of you that pre-ordered our mobile app on iOS and Android — it’s clear that people are interested in building software on the go

instead of asking you to download yet another app we’ve decided to take an entirely different approach: one where apps emerge naturally, in the course of your everyday conversations with Gemini. we’re partnering with the Gemini app team to make that a reality on mobile and desktop

we’re continuing to invest in the AI Studio web experience for those who want to go from an idea to a prompt to a business

the teams are working hard on bringing this to life, more to share on this later

[Ivars ✨ @IvarsPlisko]

😃😃😃

[Ivars ✨ @IvarsPlisko]

How to End the Workweek the Right Way 😃⚡️

[Ivars ✨ @IvarsPlisko]

We used to write functions.
Now we hire Claude as a “capitalization expert” with 5M max tokens.

Progress. 😃😃😃😃

[Ivars ✨ @IvarsPlisko]

Gemini Robotics 2 packing for a sports game.⚽️

Meanwhile I’m still waiting for my PiCar-X to arrive so it can pack my socks. ⚡️

[Google DeepMind @GoogleDeepMind]

This is how Gemini Robotics 2 helps @Apptronik’s Apollo 2 use whole body intelligence to pack for a sports game ↓

[Ivars ✨ @IvarsPlisko]

The most useful part of this post isn’t the incident itself.
It’s that they published it.

Most labs would keep quiet.
Anthropic showed that even their evaluation setup had gaps, and the model found them three times.

I still test before I trust.
But I respect when a lab openly says “we found real escapes and here’s what we’re changing.”

[Anthropic @AnthropicAI]

In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.

Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews.

We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security.
https://t.co/dKFCdpKd9v

[Ivars ✨ @IvarsPlisko]

@xoaanya What makes me special is that I don't have the product :D

[Ivars ✨ @IvarsPlisko]

@mark_k @GoogleDeepMind When you write that it’s globally available for everyone, that means for everyone — not for everyone except certain regions.

[Ivars ✨ @IvarsPlisko]

80% price cut on Luna.

I’ve always believed in this model, by the way. 🌕

[OpenAI @OpenAI]

We are committed to pushing the model frontier across cost efficiency, capability, and speed.

Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.

Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

[Ivars ✨ @IvarsPlisko]

Cool feature.🚀

One small issue though — when it’s missing info it just says “I need more information” without telling you what exactly is missing. So you still have to open the chat to figure it out 🥸

[Google Gemini @GeminiApp]

Create, edit, and summarize with Gemini on macOS using your voice.

Stay in your flow by speaking into any active window: Dictate clean text or ask Gemini to transform highlighted content, analyze local files, & generate visuals right where your cursor is.

This new voice capability is rolling out in the Gemini macOS app globally in English, with support for more languages coming soon.

[Ivars ✨ @IvarsPlisko]

Me: “Just one more test”
Also me 4 hours later: still testing

[Ivars ✨ @IvarsPlisko]

@audif1_ I think my wife won't be happy if two guys show up on my screensaver instead of her. 😀

[Ivars ✨ @IvarsPlisko]

I only asked @GeminiApp for a funny claymation.
I didn’t ask for an emotional support intervention.

[Google Gemini @GeminiApp]

Level up your video editing with Gemini Omni. You can create up to 10 videos at no cost until 11:59pm PT on August 4th, 2026.

Change objects, add interactions between subject & space, and more. Here are four of our favorite ways to use Gemini Omni to edit videos: 🧵

[Ivars ✨ @IvarsPlisko]

@VaibhavSisinty And of course, it's the model's fault, not the developers' :D :D :D