Sahibzada Allahyar

AI Agent Researcher @ @fastinoAI , Theoretical Physics @ @Cambridge_Uni , 3,000+ GitHub Stars ⭐️, Math / Physics Olympiad, Genius

[Sahibzada Allahyar @singularity_sah]

After our viral demo comparing Jev to GLiNER for browser use, there was a lot of demand for the local open source browser use so here is the GitHub repo https://t.co/lJyz1WQZPS

[Sahibzada Allahyar @singularity_sah]

36x cheaper Browser Use w/ GLiNER2.5 vs. Jev. And it's local.

[Sahibzada Allahyar @singularity_sah]

@AbsurdistBanana 🤣🤣🤣

[Sahibzada Allahyar @singularity_sah]

My browser use agent using open Jev (GLiNER) got me featured on front page news. in the biggest ai newsletter (700k+ members) 😳

[Sahibzada Allahyar @singularity_sah]

36x cheaper Browser Use w/ GLiNER2.5 vs. Jev. And it's local.

[Sahibzada Allahyar @singularity_sah]

If @DhruvAtreja1 says it, you gotta believe it. 😼

[Dhruv Atreja @DhruvAtreja1]

@singularity_sah is the guy to follow if you wanna see some cracked open source, local AI projects 🧠

[Dhruv Atreja @DhruvAtreja1]

@singularity_sah is the guy to follow if you wanna see some cracked open source, local AI projects 🧠

[Sahibzada Allahyar @singularity_sah]

@DhruvAtreja1 Much love brother 🙌❤️ wouldn’t be here without your support

[Sahibzada Allahyar @singularity_sah]

@george_onx @DhruvAtreja1 💯🙏

[Sahibzada Allahyar @singularity_sah]

If you care about local models this is a HUGE breakthrough. It’s like Claude Opus 4.5 or Qwen 3.6 running on your iPhone.

Imagine, with Phone Use task-specific models (which we are working on) you could do every tedious phone task autonomously. These models can literally use your phone, pressing buttons and messaging people for you etc.

Task specific and local was our whole thesis for making @fastinoAI

[Adrien Grondin @adrgrondin]

The iPhone 18 Pro is much better than what I expected for on-device AI

It can run a 27B model at double the speed of the 17 Pro

The new A20 Pro chip is a beast

[Sahibzada Allahyar @singularity_sah]

@ItsCuthulhu we'll take a look at this and ill speak to the creator of GLiNER about it

[Sahibzada Allahyar @singularity_sah]

Jev is INSANE for browser use 🤯

Jev ($0.0055) vs PrismML Bonsai 2 27B ($0.04) on browser use

Jev did better than Google Gemma 4 🤯

[Sahibzada Allahyar @singularity_sah]

@george_onx 💯💯

[Sahibzada Allahyar @singularity_sah]

@george_onx @MKhordoo @w0lframk0 @PatronusBen 🔥🫡

[Sahibzada Allahyar @singularity_sah]

I will be fine tuning a lightweight open source model for fully local and free Browser Use, Computer Use and Phone Use. Follow along.

[Sahibzada Allahyar @singularity_sah]

Have we achieved Claude Opus 4.6 intelligence locally running on an 8GB MacBook laptop with PrismML's Bonsai?

Here I benchmark Bonsai 2 27B vs Google Gemma on Browser Use tasks.

Should I benchmark Jev next?
https://t.co/c6hNg39B3m

[PrismML @PrismML]

Today, we’re announcing Ternary Bonsai 2 27B.

Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.

Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.

Ternary Bonsai 2 27B is available today under Apache 2.0.

[Sahibzada Allahyar @singularity_sah]

@SemiAnalysis_ son im actually crine

[Sahibzada Allahyar @singularity_sah]

@aisearchio Nope. What I attached below is the closest thing we have to open source Jev - I prove it in the thread and any serious ai scientist will agree: https://t.co/nNfkC43KUa

[Sahibzada Allahyar @singularity_sah]

I keep seeing talks about Open Jev this and that. 🤦🏻‍♂️ Grifers are grifting. The only Open Source Jev is GLiNER and every serious AI scientist will tell you this. 🧵 Let me walk you through it...
https://t.co/PBT9sP7f6X

[Sahibzada Allahyar @singularity_sah]

Exhibit F: https://t.co/FwrEG5hWlg

[Dhruv Atreja @DhruvAtreja1]

Seeing a lot of “who’s building open-jev” posts 🤦‍♂️

It has already existed before Jev, it’s fully open and fine-tuneable, has millions of downloads on hf, and is used by teams at Nvidia etc.

It’s called GliNER and @urchadeDS created it. It’s very well known in the ML community

[Sahibzada Allahyar @singularity_sah]

I keep seeing talks about Open Jev this and that. 🤦🏻‍♂️ Grifers are grifting. The only Open Source Jev is GLiNER and every serious AI scientist will tell you this. 🧵 Let me walk you through it...
https://t.co/PBT9sP7f6X

[Sahibzada Allahyar @singularity_sah]

We used Open Jev to prevent the gang of rogue OpenAI agents from hacking HuggingFace. Watch with audio 🔊.
https://t.co/7v1kfUBAYh

[Sahibzada Allahyar @singularity_sah]

Exhibit E: https://t.co/0bfchHIvTh

[Trent Zock-Robbins @tazr_dev]

Yesterday, GLiNER2.5 played doom. Rapid play looks coherent, but Jev has a secret sauce: decisions.

So, I kept the embedder and trained a small decision head on the output.

Now it's a homebrew decider model on embeddings.

What homebrew decider model will you make this weekend?