Ajay Malik

90% of AeXIV research are from China alone, they're not just building AI they're understanding it amking better harness, benchmark and more.

Gemini 3.1 Flash Lite is no longer in preview

OpenCode is Best You can connect Claude, OpenAI, Antigravity and use them all in once place

A year ago, we had Sonnet 4.2 as the top SOTA model and people were amazed and happy. Now, we have DeepSeek-V4-Flash more powerful than Sonnet 4.2 yet people are crying for GPT-5.5 and Claude 4.7.

@ai_for_success Twitter Promote them and pay them if you engage with them just block and move on.

I try to stay calm and respectful, even toward people leaving hate comments. But there is a limit, and if you cross it, don’t expect me to stay nice.

Hi SEO folks, do backlinks still matter?

@_mohansolo Why don't you provide option to plan ans execute with different models

We heard concerns that Antigravity consumes many tokens for simple tasks now. So, we're adding Gemini 3.5 Flash (Low) as a way to optimize token usage for these tasks. In our internal testing, it generates around 45% fewer tokens than Gemini 3.5 Flash (Medium) and generally outperforms Gemini 3 Flash (High) on SWE tasks. We've also gone ahead and reset Gemini quota across all paid plans to make sure you have all the tokens needed to build for the next week 🙂

AI costs keep going to rise as it gets intelligent, I don't think Aai can replace engineer when it Cost more than engineer. Engineer+ Cheap open-source AI is way forward.

Is it only me or wired earphones making a comeback

Gemini 3.5 Flash is completely Unusable, it can't follow basic instruction or do basic task which GPT-3 could do 4 year back.

@jukan05 1.4nm that too with DUV, but it still can't beat EUV in production

HUAWEI SAYS IT HAS COME UP WITH A NEW PATHWAY TO SHORTEN ITS GAP WITH INDUSTRY LEADER TSMC, POTENTIALLY ACHIEVING A BREAKTHROUGH IN MAKING ADVANCED SEMICONDUCTORS WITHOUT CUTTING-EDGE EQUIPMENT (PEOPLE’S DAILY) HUAWEI: THE KIRIN SMARTPHONE CHIP TO BE LAUNCHED THIS FALL WILL SEE A SIGNIFICANT PERFORMANCE IMPROVEMENT. (PEOPLE’S DAILY)

I like how companies find out that using AI is more expensive than hiring humans now.

@hqmank Codex shim?

DeepSeek V4 Pro and V4 Flash now running in Codex app, both using DeepSeek's official API key. Good backup when your Codex usage runs out.

@bcherny I was, I don't use it anymore.

@voy_aj 👋 are you on the latest version?

WTF Claude code is this some ester egg or something?

@oguzydz @bcherny Same here x.com/i/status/20562…

Hi @bcherny, while using Claude Code in the VS Code terminal, I’ve noticed that the terminal has occasionally been breaking/glitching over the past few days. What could be causing this?

I never found openclaw or hermes usefull but Let me see how it can help me

If I saw the AI ad of your product I ain't buying it

@kevinhou22 @OfficialLoganK This is IDE it doesn't have a /browser and even in Antigravity I used /browser it didn't used at all like its not exist

@voy_aj @OfficialLoganK Hey Ajay, browser is now hidden behind /browser. Can you try that please? We’ll work on making this more visible and/or automatic

@OfficialLoganK If you vibe code antigravity with your shitty Model this is what happen. Despite Browser tool enabled it can't access it at all

Its 2026 @deepseek_ai Chat still looks same as 2024. No agents, No coding plan, No Tools, No deep Researcher, No Plugins. Guys just give me access I'll add them all for free it barely take 1hr for such simple integration.

@morew4rd What's make it different?

@voy_aj pi.dev it's a coding agent harness similar to claude code and codex cli. you can use almost any LLM with it.

Kimi 2.6 + pi can work on a task for several hours on its own (if primed right) and fix issues. Zero extensions, raw pi + "medium thinking." Surely much cheaper than codex + /goal too.