Lentils
@Lentils80
Early AI news, leaks and cool tests.
Reach me via DMs or Lentils80@proton.me
Discord server:
120 Following 12.2K Followers
Here is a comparison between Gemini 3.8 Flash and Gemini 4 Hydrogen (the new Flash model). The results are not even close.
Additionally, Hydrogen will launch with a new 4th thinking effort: `xhigh`.
1st video: Gemini 4 Hydrogen
2nd video: Gemini 3.8 Flash
Show more
From what I've tested and seen, Gemini 4 Hydrogen seems to be a distill of Gemini 4 Carbon specifically.
It's a fast model, inherits Carbon's mannerisms (which are much better than Argon's imo) and rivals Argon at a much lower cost.
It will be a good Flash model definitely.
Show more
From what I've tested and seen, Gemini 4 Hydrogen seems to be a distill of Gemini 4 Carbon specifically.
It's a fast model, inherits Carbon's mannerisms (which are much better than Argon's imo) and rivals Argon at a much lower cost.
It will be a good Flash model definitely.
Show more
Gemini 4 Hydrogen (gemini-4-hydrogen) private EAP officially started. Release is expected soon™
3D capability isn’t the only area where Gemini 4 Carbon and Krypton massively improve upon Gemini 4 Argon.
Compared to Argon, the frontend and UI on Krypton are much better and more tasteful, and it's also more creative.
Show more
Gemini is improving fast lately, seems like they're now aiming for a major model release every month or so
1st video: Gemini Krypton (very early checkpoint)
2nd video: Gemini 4 Carbon
3rd video: Gemini 4 Argon
Show more
Gemini is improving fast lately, seems like they're now aiming for a major model release every month or so
1st video: Gemini Krypton (very early checkpoint)
2nd video: Gemini 4 Carbon
3rd video: Gemini 4 Argon
Show more
When will AI labs learn that deliberately handicapping models in cybersec (which is literally just coding for the most part) will generalize to other coding areas as well in real world tasks? Also, it will make it worse at defensive cybersec. There's a reason Opus 4.7 was hated
ofc not saying this is a bad thing for the world (it's not) considering recent events, but there still has to be another better way to do this
Show more
An output from the updated Carbon checkpoint 😼
Like I said in my previous post, Carbon is noticeably better than Gemini 4 Argon at 3D (tho still not perfect). It seems like Google is currently focusing on improving that domain before moving on to frontend/UI
Show more
The Carbon checkpoint on LMArena and Gemini Web/App has been updated.
Gemini 4 Carbon EAP has officially started. Early access has been given to partners and it's being tested in LMArena. The release is very close.
It's noticeably better than Gemini 4 Argon at 3D, though frontend/UI is still bad. It's also slower but less lazy.
Show more
The 40th internal model leak this month from OpenAI themselves. Impressive.
These showcases sure look familiar 🙀
I saw a bunch of Muse Wonder outputs from multiple people. This model is such a hot steaming piece of benchmaxxed garbage I couldn't even believe my eyes.
Not even Gemini models are this benchmaxxed bro.
Show more
muse wonder vs swe 2
tested both models with the same prompt at highest reasoning available but came out really different
> muse took 24 mins and cost about $18
> swe took 15 mins and cost zero dollars
really surprised by how good swe 2 did here but i don't know what muse wonder was doing
model is scored above astra and fable 5.1 but it’s performing way worse than both in tests
Show more
Codex has been completely broken the past couple of days with these reconnecting errors; pauses my work every hour or so making it impossible to do anything.
It's not just me either, plenty of people having the same issues.
@thsottiaux please fix.
Show more
A new GPT memory consolidation model, called Memory4 Dream (ID: memory4-dream), has appeared a few hours ago and it seems to be based on GPT-6 Luna.
interesting 🤔
They really just reused the internal codename for the public release of Gemini 4 Argon lmao. Well at least that's how you know my info was accurate 😛
ig the main difference they made since then is upping the <output limit> from 256k to 1M tokens
Show more
A new Gemini 4 Pro checkpoint dropped and it is just another time waste. If Google doesn’t bring back Ultra (large sized models vs current medium sized Pro models) it’s already over for them. Keep in mind that all Google releases are already nerfed on Day 1.
Show more
New Gemini 4 Pro checkpoint w/ extended thinking on
Expected more but this seems all DeepMind can do. Gemini 4 prepped for release in weeks (seen internally). Vs Opus 5.5 it's a gnat but isn't the Gemini 4 to wait for. Release version may be better/worse
Show more
OpenAI is getting demolished wtf??
Lots of users are already being routed to Claude Sonnet 5.5, even on free tier.
Stop basing your release date "predictions" on Polymarket odds bro, it's too obvious now
Google is also actively working on model safety, especially Frontier and CBRNE risks (YOLO indeed), aiming to become more like OpenAI and Anthropic.
peep those "teamfood" and "dogfood" labels though, iykyk 👀
Show more
HOLY MOTHER OF MATHEMATICS!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Google is working on a math-focused variant of its DeepThink model, and its raw thoughts are pretty funny.
Gemini 4 Pro checkpoints have finally started appearing internally a few days ago.
This is the first ever output from the model, internally codenamed "argon". It took 2.4 minutes on High thinking effort.
It has a 256k token "output limit", compared to 64k in previous Gemini models. I also heard it will ship with a 2M context window, tho it's still not decided if they'll do it.
Show more
