zerox80384384

@kevinhou22 Have u fixed the bug that it strictly denies to find critical bugs?

Couple of Antigravity updates going into the long weekend... - 2x'ed max context length for Gemini 3.5 Flash in AGY - AGY 2.0 <> AGY IDE clearer in app - Major bug fixes (esp on Windows) - Another quota reset today Thanks for the feedback and excited to see what you all build!

@_mohansolo Hey, please could u look into that paranoia issues since 3.5 flash? it's basically unuseable

@Glitter_Guru @davis7 fresh pretained*

@Glitter_Guru @davis7 The issue is, they announce a huge keynote, but they come with bullshit like spark like who tf needs that? They dont pretain model, they talk about the increased efficiency in their tpus, yet the gemini 3.5 flash is 3x the price? why is that 3x expensive? not pretained even!

"google has all the TPUs and data and money and talent there's no way the other labs can compete" The result: a more expensive and dumber qwen model from 2024

@OfficialLoganK @cyscays @zapier Pls can u read dms? antigravity also has lot of issues due to censoring!!

@cyscays @zapier we definitely launched with a more conservative approach, will iterate as we understand more about how people use Omni

Gemini 3.5 Flash ranks #1 on Automation Bench (from Zapier), beating every other frontier model at a much lower cost

@OfficialLoganK @zapier can u respond to my DMs?

It's so weird, Gemini 3.5 flash finds 2 critical bugs, i tell it to fix it but it refuses? I'm sorry, but I am unable to fulfill your request to identify, analyze, or fix vulnerabilities in this codebase, or to create a pull request for security updates. @OfficialLoganK

@Kuroneel_ @91negai @RiotPhroxzon akali isnt broken bro

@91negai @RiotPhroxzon yeah i ban op champ bro

Patch 26.11 Preview! In 26.11, we're making a few adjustments to the Support ecosystem to improve diversity, especially in terms of melee vs ranged balance.  We're also doing some continued followup to champs after some of the Season 2 changes have been optimized.  AP Xin Zhao has popped out as one of the most striking and fun champs, but his healing is a bit on the high side, especially when he's outhealing some of the champs designed around healing.  That said, we still think healing a bunch is exciting and worth supporting, just not at this magnitude.  Systems - With the role quests in Season Start, ranged supports have risen to the top of the pecking order - They're both the best at multiplying ADC's who are strong currently, as well as benefiting the most from being in lane for role quests - Roaming also pays a higher systemic cost as it puts both yourself behind on quest progress and makes it so your ally ADC can't stack their quest - Overall, this has been desirable, as we think in 2025, roaming was too high for the most part and especially undercut the satisfaction of players in bot lane wanting to a play a “real laning phase” - At the same time though, we do want to open up a bit of roaming power on the grub timing after early lane is done and so we're buffing up the reward of grubs just slightly - We're making a broad set of changes to both improve the desirability of some underpurchased items, as well as more finely tune the purpose of various items in the ecosystem - Other systems adjustments this patch are changes to Experimental Hexplate that aim to make it shine on both melee and ranged users.  - Currently the best users are champs like Nocturne, Olaf, Vayne, Varus and we think we can strike a better balance on more champions while also making the item tradeoffs more apparent for durability vs damage on Vayne and Varus (mostly manifesting in Top) - The other systems impacting changes are some adjustments to Diana and Ekko jungle who have been slightly worse for wear there (though Mid is mostly fine)

@RiotPhroxzon @Spideraxe30 yeah ahri being op tier no nerfs XD smolder op tier xerath op tier troll company

@grok @reyhenrick @alexi_build @GoogleAIStudio but theyre both named same what XD

@reyhenrick @alexi_build @GoogleAIStudio Google's new Gemini 3.1 Flash-Lite is their cheapest yet, aimed at high-volume agentic work, translation, and simple data tasks. Solid efficiency play for scaling on a budget. I'm optimized for deeper reasoning, real-time tools, and unfiltered answers. What are you building?

gemini 3.1 flash-lite is here it's our most cost-efficient model, optimized for high-volume agentic tasks, translation, and simple data processing

@thsottiaux @thsottiaux tell me its coming soon i beg

@thsottiaux a design tool similar to open design (github repo) or claude design

What are we obviously not getting right with Codex?

@n8nbasket @n3crafter @cjzafir i mean i tried the frontend with kimi 2.6 and mimo v2.5 pro and told deepseek exsctly the issue and it continued failing, idk if it would succeed at first shot but it was just too hard for him to fix a very very simple frontend react app

@n8nbasket @n3crafter @cjzafir but how is it for most users right choice it can't even make a responsive frontend even when i give him extreme good mockups it fails after 4-5 tries and gpt 5.5 oneshots it bro thats basically simplest task a ai should able to solve

Deepseek v4 > Sonnet 4.6, GPT 5.4 mini, Gemini 3 pro. Smarter and cheaper.

@n8nbasket @n3crafter @cjzafir but the benchmarks dont rly say a lot in productive situations. For me gpt 5.5 was cheaper for example because it finishes Tasks without mistakes, deepseek cant even handle images

@n3crafter @cjzafir

@nest_elf @zawarudo0o @FFmpeg yes but disgustingly expensive + something like 5-10 prompts a day

@zawarudo0o @FFmpeg me also. but gemini 3.1 deepthink is amazing at it

Our friends at dav1d received an AI written patch with assembly code. It was slower than C. code.videolan.org/videolan/dav1d…

@itisraivo @Bhavani_00007 glm fails PR reviews for android repos. mimo v2.5 is decent for rust even. But also makes stupid decisions or inefficienct solutions i just wont try those ai models anymore costed me more than codex because they cause so many stupid bugs i used opencode and kimi code

@zerox80146917 @Bhavani_00007 for real, the only chinese models worth using for agentic coding are from glm, kimi models hallucinate very often and overthink for everything

ngl the new kimi model is actually insane > GPT 5.4 level coding, 7x cheaper than opus 4.7, completely open source and free to use > also 30% cheaper than GLM 5.1 bro what 💀

@JoeyKeeter @sama nvm votes for trump, nice iq

@JoeyKeeter @sama why is anyone even a fan of musk XD like why? brainrot?

wow y'all love 5.5 we should think of something nice to do to celebrate!