qilua

AI, Tech, part time troll

[qilua @qilua02]

lost my blue check now, the 50% discount for 2 months is gone too since i'm out of balance

but weirdly my recent posts are getting more views than when i had the checkmark

gonna keep posting content anyway without it, just to see how the progress looks without blue check

[qilua @qilua02]

I only have regular X Premium since it's all I can afford right now, not Premium+. I'm currently on a 50% discount for the first two months as a new user, and this is only my first month. Grok from Premium is only accessible through the website, so I can't use it with local AI agents.

Tools like Hermes Agent need to be installed locally on my computer and require an API key or AI subscription that supports OAuth login for local agents. Right now, regular X Premium's Grok access doesn't support OAuth for tools like Hermes Agent only SuperGrok or Premium+ does.

That's why I'm posting a lot about AI and tech right now, including some trolling, trying to grow fast and earn revenue sharing before my discount ends. The goal is to use that money to buy a real AI subscription I can actually use with Hermes Agent, OpenCode, and other agent

[qilua @qilua02]

@STUD_MAN_X Mostly for a personal side project and internal tooling. Rewriting around 500k lines of code consumes a massive amount of tokens and time.

[qilua @qilua02]

I got $33 free credits for ChatGPT on Codex and Claude Code. Here is how to get it 👇

[qilua @qilua02]

Claim the $33 credits here 👇 https://t.co/J42GMbB2w5

[qilua @qilua02]

Setup guide for Codex CLI 👇 https://t.co/ZENuKDnQBQ

[qilua @qilua02]

@limtzugar Appreciate the info, thanks!

[qilua @qilua02]

Wait, another ~50M tokens for free from Cline?! 🤯

You get 4 models out of the box:

- deepseek-v4-flash
- GLM-5.2
- step-3.7-flash
- laguna-s-2.1

No API key or credit card needed!

Install: npm i -g cline
Website: https://t.co/lpQJlICRBm

[qilua @qilua02]

DeepSeek-V4-Flash-0731 is out on Hugging Face along with its technical report.

A 304B model that performs on par with GLM 5.2 753B.

Link: https://t.co/uCuXDkI6bt

[qilua @qilua02]

Wait what?! This new DeepSeek v4 Flash update is literally catching up to GLM 5.2, this jump is crazy 😭

[qilua @qilua02]

~100M free tokens with the new DeepSeek V4 Flash 0731! 🤯

Link: https://t.co/m5L69SU3tM

Grab it before it’s gone.

[qilua @qilua02]

Wait what?! This new DeepSeek v4 Flash update is literally catching up to GLM 5.2, this jump is crazy 😭

[qilua @qilua02]

@tranletrongvan Yeah for sure! You can just install the Cline extension directly in Cursor from the extensions tab.

[qilua @qilua02]

@tranletrongvan Feels super fast, my guess is around 40+ tokens/s. For the exact rate limit, I'm not really sure since they didn't explicitly mention it.

[qilua @qilua02]

Wait what?! This new DeepSeek v4 Flash update is literally catching up to GLM 5.2, this jump is crazy 😭

[qilua @qilua02]

deepseek-v4-flash is honestly enough for my daily use, while glm-5.2 handles my complex stuff.

Plus, they’re 3x to 10x cheaper than US models with the same level of intelligence.

Just leaving my note here as of July 2026.

[qilua @qilua02]

@deepseek_ai wow

[qilua @qilua02]

@MiniMax_AI @Hailuo_AI wow

[qilua @qilua02]

@STUD_MAN_X Not really sure tbh, I’m still on free X so I can’t see the analytics tab.

My guess is probably around 500k in the last 2 months.

[qilua @qilua02]

I hit 500 followers today, July 31, 2026.

Next target: 5M impressions.

[qilua @qilua02]

Hit 400+ followers on July 30, 2026.

Posting this to track my progress and see how much this account grows over time.

[qilua @qilua02]

@AbionMorse Mimo 2.5 still feels wasteful to me since smaller vision models fail to follow long prompts properly anyway.

If I actually need vision or heavy file analysis, I’d rather just use flagship/top-tier models. Saves me the frustration of fixing bad outputs.

[qilua @qilua02]

deepseek-v4-flash is honestly enough for my daily use, while glm-5.2 handles my complex stuff.

Plus, they’re 3x to 10x cheaper than US models with the same level of intelligence.

Just leaving my note here as of July 2026.

[qilua @qilua02]

@Christian_22023 90% of my work is pure text though. Even with discount, 5.6 Luna’s output is still ~3x more expensive ($0.60 vs $0.18).

Paying for vision/multimodal features I don't use is a waste. Text-only models + MCP and detailed prompt engineering do the job just fine for way cheaper.

[qilua @qilua02]

@winglee nice

[qilua @qilua02]

Burned over 600M tokens in a single day and got all of it for FREE. 🔥

Realistically, if I had to pay for this, I could never afford this many tokens. Wild stuff.

[qilua @qilua02]

You can get over 1 Billion FREE tokens for DeepSeek V4 Flash, MiMo 2.5, Nemotron 3 Ultra, and Ling 3.0 Flash.

I've been using this on Zcode (https://t.co/328dpgDTlT) for over a month now.

Here's how to set it up 👇