Sabbir Hossain

CTO @velox_studio_26 • AI/ML Dev • Trader • Turning ideas into products, GameDev (Hobby)

Building The next gen of education system

[Sabbir Hossain @Sabbirbyte]

@TrumpTruthOnX Stop poking around and you will be good

[Sabbir Hossain @Sabbirbyte]

@louszbd So the upcoming glm model will fix this?

[Sabbir Hossain @Sabbirbyte]

@ziwenxu_ does it work properly for toolcalls and other usecase? because for me sometime the toolcalls and chat have some parameter/payload issues

[Sabbir Hossain @Sabbirbyte]

🚀 Big news from Alibaba's Qwen team!

Qwen3.8 is launching soon as open-weights — a massive 2.4 trillion parameter model that's already being called one of the most powerful AIs available today. It’s positioned right up there with the frontier leaders, second only to Fable 5.

(Note: the current parameter-size leader on the board is Moonshot’s Kimi K3 at 2.8T — the scale race is heating up fast!)

You don’t have to wait for the full release though! The Qwen3.8-Max-Preview is live right now on Alibaba’s Token Plan, Qoder, and QoderWork. Early access unlocked.

This continues Qwen’s incredible momentum as the world’s most downloaded open-source AI family. Excited to see what the community builds with it once the weights drop.

Who’s jumping in to test the preview? 👀

Links:
International Token Plan → https://t.co/cQO668RlTF
China → https://t.co/E853OeblMs

[Sabbir Hossain @Sabbirbyte]

We just saw a killer independent eval from Vercel CTO @cramforce on real cybersecurity tasks (using their open deepsec harness on vulnerable code):

• Kimi K3 = the sweet spot: excellent recall/precision at a great price
• GPT 5.6 Sol = most thorough analysis but ~7x more expensive
• GLM 5.2 = strong runner-up at even lower cost

Verdict: Use GPT 5.6 for one-time deep baseline audits, then Kimi K3 for continuous scanning. Fable 5 refused everything.

Chinese frontier models are eating the cybersec use-case. Game changer for cost-conscious teams.

Link: https://t.co/GuQfEWOfZ0

[Malte Ubl @cramforce]

We ran Kimi K3 on a private cybersecurity benchmark.

TL;DR: Kimi K3 is the workhorse for cyber security tasks at great recall/precision/price. GPT 5.6 is best recall/precision but at 7x higher cost per run.

For context, https://t.co/UMvysvNW5w is an open-source cyber harness designed for finding vulnerabilities in large codebases.

The eval runs deepsec on an undisclosed open-core application at a git sha before a large number of security issues were fixed. This is a secret eval that cannot be directly benchmark-maxxed.

S-Tier: GPT 5.6 Sol: By far the most thorough analysis, but coming in at over 7x the price of the runner up.

Best price/recall: Kimi K3. Next tier of recall at a good price

Best price at good recall: GLM 5.2 (40% lower price than Kimi K3)

GPT 5.5: Only recommended with subscription or high-discount API price. Similar recall to Kimi at much higher list price.

Opus 4.8: Only recommended with subscription or high-discount API price. Similar recall to GLM 5.2 at much higher list price.

Fable 5: 100% refusal rate. Cannot be used for security analysis.

Sol on a large code base will quickly get into 6-figure pricing. This is still affordable relative to the risk of letting security issues unfixed or paying bug bounties.

I'd recommend using Sol for a one-time baseline and then using Kimi K3 for continuous analysis.

When using open-weight models, make sure to use an inference vendor that supports zero data retention.

[Sabbir Hossain @Sabbirbyte]

@Rainmaker1973 why its keep making it fatter ?

[Sabbir Hossain @Sabbirbyte]

@cramforce so, gpt 5.5 and opus 4.8 no longer the avg using model for this tasks>

[Sabbir Hossain @Sabbirbyte]

For now, I’ve seen enough Kimi K3 posts 😂

Don’t get me wrong , it’s an absolute beast and my main daily driver together with GLM 5.2. Both are killing it.

But X algo… can we mix it up?
Show me fresh AI inventions, new model architectures, algorithm breakthroughs, crazy tech experiments, and real innovation , not just one model on repeat.

Feed me the full timeline of what’s actually moving in AI right now. Thanks 🙏

[Sabbir Hossain @Sabbirbyte]

The single biggest upgrade that made our AI learning path generator actually good:

I added a second agent whose only job is to argue with the first one.

One LLM?
→ Super confident.
→ Often wrong.
→ Sounds impressive anyway.

Two LLMs debating?
→ Catches hallucinations.
→ Stress-tests every recommendation.
→ Outputs paths people actually finish.

Before: Generic, overly optimistic plans that fell apart.
After: Battle-tested learning paths that adapt and hold up.

The debate format was the cheat code. One model alone is theater. Two fighting it out is truth.

Join the waitlist → https://t.co/ElO7faDle1

[Sabbir Hossain @Sabbirbyte]

@maxedapps it will fix soon once the reasoning_effort rolls out for different level

[Sabbir Hossain @Sabbirbyte]

@claudeai Pro users certainly dont gonna renew any more if they dont get to use fable 5

[Sabbir Hossain @Sabbirbyte]

I tested Kimi K3 across multiple complex 3D rendering and interactive game logic tasks, and the results are absolutely shocking. 🤯

From rendering complex real-time geometric physics to procedural terrain generation and interactive fluid animations, it handles it all flawlessly. Check out the video to see it in action!

[Sabbir Hossain @Sabbirbyte]

🚀 Kimi K3 reset complete — time to push it harder.

Just dropped 4 heavy 3D challenges at once:

• Highly interactive 3D scene (no GSAP, custom scroll + interpolation, proper React render-loop separation)
• Complex dynamic environment with correct depth/scale lighting + programmatic scene (no prebuilts)
• Small real-time 3D game with proper collision & entity system (no Unity-style shortcuts)
• Custom shader material from scratch (vertex + fragment logic, uniforms, math breakdown)

All in one go. Let’s see what she can actually do now.

Testing mode = activated 🔥

[Sabbir Hossain @Sabbirbyte]

@theo Kimi code works well for me

[Sabbir Hossain @Sabbirbyte]

@mweinbach The output looks clean though

[Sabbir Hossain @Sabbirbyte]

@bijanbowen Is video up? Its a great model though

[Sabbir Hossain @Sabbirbyte]

Just spent hours benchmarking Moonshot’s brand-new Kimi K3 on 3D generation and web game logic.

I skipped the standard coding benchmarks and went easy-to-hard instead: pushing its physics, UI, rendering, and real-time logic.

It delivered 3 impressive interactive projects on the first try:

🌲 A cozy low-poly Floating Island with smooth orbital camera, dynamic lighting slider (day → sunset → night), and proper shadows.

🎛️ A real-time 3D Customizer dashboard — instant mesh swaps, size/rotation sliders, color palette, and wireframe toggle.

👾 A fully functional 2D Survivor-style arena game with WASD + Space attack, dynamic swing arc, auto-spawning enemies, scaling difficulty, health bars, and retry screen.

I ran out of tokens before I could test the other 4 ideas I had lined up.

The recent price increase is absolutely acceptable — the quality jump makes it worth every penny.

Kimi K3 is next-level 🔥

[Sabbir Hossain @Sabbirbyte]

@cheatyyyy its thinks too much honestly