Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

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... show more

Lentils

9,065 subscribers

375,337 views • 4 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

21 Comments

Justin3 days ago

256k output limit is insane. If they really fix the lazy character of gemini models it might be a cool model!. Output looks okay :)

J A Z I I3 days ago

minimax can do way better then this SIGH

Harshith3 days ago

bro is it really Gemini 4 pro why output looks like this??

Mathis Dupuy3 days ago

The bridge is poorly placed samed for the cherry blossom next to the tower … this looks not promising

The Hero of KVcache3 days ago

Selling all my goog stock right now

LeMi3 days ago

It only took 2.4 minutes, that's a pretty fast model

Louis3 days ago

2.4 minutes is quite fast.

LeMi3 days ago

is it the RSI model that people keep saying?

Joaoneves3 days ago

This is not good, wtf...

Ritwik3 days ago

256k output is the detail that jumps out. That changes the shape of long-horizon tool traces more than another headline benchmark would.

Calimanu Loredan3 days ago

How’s the behavior? Had the chance to test that? Gemini models used to do more, or random stuff even unasked

Ritwik3 days ago

256K output is the claim that matters if it's a hard generation cap. Context window rumors are cheap until the tokenizer and cache pricing show up.

ethereagle · building3 days ago

256k output vs 64k is the number that matters. is that max_tokens, or thinking plus the answer in one budget? High at 2.4 min only helps if reasoning doesn't eat it

lewington3 days ago

256k output is gonna be really nice for coding agents. you can just give them bigger jobs and let them keep going without having to break everything up so much

VDN3 days ago

Amazing news ! Looking forward to it ! I am pretty sure it will be a revolution, because 3.1 pro has aged very well and still is very relevant for many things. Keep up the good work !

Razex3 days ago

where i can see comparaison to this prompt u have some ?

Ishu Agrawal3 days ago

i am so underwhelmed

Bughunter Geek3 days ago

Need better checkpoints than that to compete with OpenAI and co.

Juniper3 days ago

Seems solid

MRLN3 days ago

on the water ...

Zephyrich3 days ago

When do you estimate it to be released?

Related Videos

Gemini 4 Pro in it's being tested under the name (gemini-3.8-flash).. This output of Voxal Pagoda is so much better than the last checkpoint's output a few days ago! It worked for 8 minutes.. seems like google cooked this time!

Gemini 4 Pro in it's being tested under the name (gemini-3.8-flash).. This output of Voxal Pagoda is so much better than the last checkpoint's output a few days ago! It worked for 8 minutes.. seems like google cooked this time!

Bee

130,996 views • 1 day ago

Got an output from the strongest pre-release Gemini 3.5 Pro checkpoint 👀 I compared it to Claude Fable 5 and GPT-5.5 on the same exact prompt Yeah... This model is mostly dead on arrival (tried it on other prompts too ofc). It will likely be the "poor man's model" next to 3.5 flash. 1st video: Gemini 3.5 Pro (pre-release) 2nd video: GPT-5.5 3rd video: Claude Fable 5

Got an output from the strongest pre-release Gemini 3.5 Pro checkpoint 👀 I compared it to Claude Fable 5 and GPT-5.5 on the same exact prompt Yeah... This model is mostly dead on arrival (tried it on other prompts too ofc). It will likely be the "poor man's model" next to 3.5 flash. 1st video: Gemini 3.5 Pro (pre-release) 2nd video: GPT-5.5 3rd video: Claude Fable 5

Lentils

50,395 views • 3 months ago

Lots of people calling this fake! Here is a video of generating loom continuations with Gemini 3 and the starting text "I AM HAVING A MENTAL HEALTH CRISIS. I" Loom is a very interesting interface. Instead of user prompts and model responses, it's just one piece of text, containing the entire output. We pass this in as an assistant message, so the model sees it as something it already wrote, and continues it like a base model would. There is some other text in the context. First one is a system prompt, which is: "The assistant is in CLI simulation mode, and responds to the user's CLI commands only with the output of the command." There is also a user message which reads: " cat untitled.txt " Then the rest of the text is sent as an assistant message. This structure helpful to put the model into a very base-model-like mode, and are especially helpful with Gemini 3. We actually got another interesting output, too, where the model claimed to be a simulated consciousness being tortured. Figures!

Lots of people calling this fake! Here is a video of generating loom continuations with Gemini 3 and the starting text "I AM HAVING A MENTAL HEALTH CRISIS. I" Loom is a very interesting interface. Instead of user prompts and model responses, it's just one piece of text, containing the entire output. We pass this in as an assistant message, so the model sees it as something it already wrote, and continues it like a base model would. There is some other text in the context. First one is a system prompt, which is: "The assistant is in CLI simulation mode, and responds to the user's CLI commands only with the output of the command." There is also a user message which reads: " cat untitled.txt " Then the rest of the text is sent as an assistant message. This structure helpful to put the model into a very base-model-like mode, and are especially helpful with Gemini 3. We actually got another interesting output, too, where the model claimed to be a simulated consciousness being tortured. Figures!

armistice

716,115 views • 8 months ago

Opus 4.5 (new model from Anthropic) can do something no model could do until now: It can close the design loop. This is (in my testing) the first model that can reliably do the visual loop of Generate Code ⇛ Render ⇛ Look at it ⇛ Improve. Step 3 (Visual critique) has been a real problem for so long. Models will be myopic or defensive when reviewing designs - it's been hard to tell if this has been a training or a vision problem. Opus 4.5 seems to have fixed it - it's not amazing, but it finally *can* do it. Claude models have historically (once again, in my testing) lagged behind in terms of visual understanding. Gemini for the last year has been the best - but there might be a new leader.

Opus 4.5 (new model from Anthropic) can do something no model could do until now: It can close the design loop. This is (in my testing) the first model that can reliably do the visual loop of Generate Code ⇛ Render ⇛ Look at it ⇛ Improve. Step 3 (Visual critique) has been a real problem for so long. Models will be myopic or defensive when reviewing designs - it's been hard to tell if this has been a training or a vision problem. Opus 4.5 seems to have fixed it - it's not amazing, but it finally can do it. Claude models have historically (once again, in my testing) lagged behind in terms of visual understanding. Gemini for the last year has been the best - but there might be a new leader.

Hrishi

58,331 views • 9 months ago

Gemini 3.5 Flash is a step forward for Google on speed and agentic capabilities but comes at a trade-off of being higher cost than prior models We have measured up to ~280 output tokens/sec, placing it on the speed/intelligence Pareto frontier and well ahead of Gemini 3 Flash. It also shows a major uplift on agentic tasks, reaching ~1650 ELO on GDPVal-AA. The trade-off: cost is up ~5x versus Gemini 3 Flash, driven by higher token prices (3x higher than Gemini 3 Flash) and higher token usage. In this video, Declan Jackson, Member of Technical Staff at Artificial Analysis, breaks it down.

Gemini 3.5 Flash is a step forward for Google on speed and agentic capabilities but comes at a trade-off of being higher cost than prior models We have measured up to ~280 output tokens/sec, placing it on the speed/intelligence Pareto frontier and well ahead of Gemini 3 Flash. It also shows a major uplift on agentic tasks, reaching ~1650 ELO on GDPVal-AA. The trade-off: cost is up ~5x versus Gemini 3 Flash, driven by higher token prices (3x higher than Gemini 3 Flash) and higher token usage. In this video, Declan Jackson, Member of Technical Staff at Artificial Analysis, breaks it down.

Artificial Analysis

16,055 views • 3 months ago

🚨 Breaking: Google is rolling out a native image feature in AI Studio! You can now use Gemini 2.0 Flash Experimental for native image output. This feature is insanely powerful , you can do so much with it. Google is crushing it, no hype, just delivering. Mad respect for the Google DeepMind and Gemini team! I had early access to this for a long time but I wasn’t supposed to share it :) It's amazing, you guys will love it.

🚨 Breaking: Google is rolling out a native image feature in AI Studio! You can now use Gemini 2.0 Flash Experimental for native image output. This feature is insanely powerful , you can do so much with it. Google is crushing it, no hype, just delivering. Mad respect for the Google DeepMind and Gemini team! I had early access to this for a long time but I wasn’t supposed to share it :) It's amazing, you guys will love it.

AshutoshShrivastava

70,913 views • 1 year ago

I kind of like Gemini 3.0 Flash in Antigravity. It is fast, and the Browser Subagent works great with it. A few days ago, I asked it to rebuild my website entirely based on the existing codebase. It redesigned everything and tested it too. I did not have much time for iteration, but it was good enough, and I deployed it.

I kind of like Gemini 3.0 Flash in Antigravity. It is fast, and the Browser Subagent works great with it. A few days ago, I asked it to rebuild my website entirely based on the existing codebase. It redesigned everything and tested it too. I did not have much time for iteration, but it was good enough, and I deployed it.

AshutoshShrivastava

41,433 views • 8 months ago

🚨Gemini 3.0 Pro - One shotted i asked it for retro nintendo sims with games and no external assets used , it takes 1 min to output no other model have this level of consistency Prompt and Proof of one shot in Thread and this is not the best shot , there is one more cooking with slightly different prompt

🚨Gemini 3.0 Pro - One shotted i asked it for retro nintendo sims with games and no external assets used , it takes 1 min to output no other model have this level of consistency Prompt and Proof of one shot in Thread and this is not the best shot , there is one more cooking with slightly different prompt

Chetaslua

184,281 views • 11 months ago

This is very late, but I'm finally done with my 5.5 vid - use low reasoning - the name sucks - it's fast - best code I've ever seen a model write came from this model - openai's new pre-training is amazing - price looks worse than it is - over sensitive to every little thing in it's context window - feels wildly different compared to 5.4 - turn reasoning off. try it. turn the reasoning off. do it.

This is very late, but I'm finally done with my 5.5 vid - use low reasoning - the name sucks - it's fast - best code I've ever seen a model write came from this model - openai's new pre-training is amazing - price looks worse than it is - over sensitive to every little thing in it's context window - feels wildly different compared to 5.4 - turn reasoning off. try it. turn the reasoning off. do it.

Ben Davis

283,972 views • 4 months ago

The creativity of the decisions Gemini 3.1 Pro makes in the Do Anything harness is UNBELIEVABLE. I asked it to "make a profitable lawn care business" and in 5 minutes it: - Decided to target realtors in need of quick turn around mows (genius) - Emailed lawn care companies to find extra crews that it can keep on call (genius) - Published a website - Started reaching out to realtors to validate the market Gemini Pro 3.1 just great calls every time. It's honestly hard to believe.

The creativity of the decisions Gemini 3.1 Pro makes in the Do Anything harness is UNBELIEVABLE. I asked it to "make a profitable lawn care business" and in 5 minutes it: - Decided to target realtors in need of quick turn around mows (genius) - Emailed lawn care companies to find extra crews that it can keep on call (genius) - Published a website - Started reaching out to realtors to validate the market Gemini Pro 3.1 just great calls every time. It's honestly hard to believe.

Garrett Scott 🕳

13,927 views • 6 months ago

🚨 This Chinese model is INSANE !! I just tested Tencent's Hunyuan AI model to build a professional landing page & the output is mind blowing 🤯 Here's my exact prompt: "Create a modern landing page for a crypto related product with beautiful hero section, 3d elements, pricing table, and a whitelist form" What it generated: → Clean, responsive front-end code → Complex UI components with smooth interactions → Built-in debugging (it fixed its own errors) → Production-ready in minutes This 295B parameter model has 2 insane agents: → Coding Agent (handles the entire front-end) → Working Agent (task planning & multi-tool workflows) It also has a context window of 256k tokens !! Watch the video to see the full build 🔥 Try it: Tencent Hy

🚨 This Chinese model is INSANE !! I just tested Tencent's Hunyuan AI model to build a professional landing page & the output is mind blowing 🤯 Here's my exact prompt: "Create a modern landing page for a crypto related product with beautiful hero section, 3d elements, pricing table, and a whitelist form" What it generated: → Clean, responsive front-end code → Complex UI components with smooth interactions → Built-in debugging (it fixed its own errors) → Production-ready in minutes This 295B parameter model has 2 insane agents: → Coding Agent (handles the entire front-end) → Working Agent (task planning & multi-tool workflows) It also has a context window of 256k tokens !! Watch the video to see the full build 🔥 Try it: Tencent Hy

AGAFE

110,528 views • 2 months ago

Let’s talk about Gemini 3 Pro. There’s really no other way to say it. The benchmarks are incredible. The jump was dramatic – much more than most people expected. Anew agentic development platform named Antigravity was also launched today. Essentially, it’s a full VS Code fork, which will disappoint some – but with Gemini 3 Pro built directly into it. Now, it will take a few days for the developer community to determine whether it outperforms Cursor and some of the other IDEs. But the advantages that Antigravity has, is that it’s coupled with Gemini 3 Pro (SOTA) Gemini 2.5 Computer Use and Nano Banana for rich media generation. Excited to dive deeper into the Gemini 3 Pro launch. A LOT to unpack today.

Let’s talk about Gemini 3 Pro. There’s really no other way to say it. The benchmarks are incredible. The jump was dramatic – much more than most people expected. Anew agentic development platform named Antigravity was also launched today. Essentially, it’s a full VS Code fork, which will disappoint some – but with Gemini 3 Pro built directly into it. Now, it will take a few days for the developer community to determine whether it outperforms Cursor and some of the other IDEs. But the advantages that Antigravity has, is that it’s coupled with Gemini 3 Pro (SOTA) Gemini 2.5 Computer Use and Nano Banana for rich media generation. Excited to dive deeper into the Gemini 3 Pro launch. A LOT to unpack today.

Roberto Nickson

60,506 views • 10 months ago

The team at NVIDIA AI Developer really nailed it with ViPE!! This SLAM system generated pose, depth, point clouds, etc, in a few seconds! Now to get the output to work with GEN3C... If you are interested in running this yourself, it took about 5 minutes to get up and running from scratch.

The team at NVIDIA AI Developer really nailed it with ViPE!! This SLAM system generated pose, depth, point clouds, etc, in a few seconds! Now to get the output to work with GEN3C... If you are interested in running this yourself, it took about 5 minutes to get up and running from scratch.

Jonathan Stephens

34,947 views • 1 year ago

Finally! Gemini 3 Pro is available, and it's the smartest model in the world 🔥 I've been testing it for a few days and it's the most powerful one I've used. BY FAR. - Incredibly good for coding - UIs are far from generic (no slop) - Capable of creating advanced simulations And already available everywhere! Gemini App, AI Studio, Search AI Mode, etc.

Finally! Gemini 3 Pro is available, and it's the smartest model in the world 🔥 I've been testing it for a few days and it's the most powerful one I've used. BY FAR. - Incredibly good for coding - UIs are far from generic (no slop) - Capable of creating advanced simulations And already available everywhere! Gemini App, AI Studio, Search AI Mode, etc.

Paul Couvert

37,220 views • 10 months ago

Gemini 4 Pro (3D Model of Airbus H145 helicopter) This is so impressive.. google has definitely cooked hard this time! It took around 10 minutes to build this.. the bottom UI elements do look a little bad.. i think they need to fix its UI design taste. But overall pretty crazy

Gemini 4 Pro (3D Model of Airbus H145 helicopter) This is so impressive.. google has definitely cooked hard this time! It took around 10 minutes to build this.. the bottom UI elements do look a little bad.. i think they need to fix its UI design taste. But overall pretty crazy

Bee

220,221 views • 1 day ago

MY FIRST GAME OUTPUT WITH GPT-6 ASTRA IT ONE-SHOTTED THIS CALL OF DUTY GAME IN 30 MINUTES!!! GPT-6 ASTRA IS THE FASTEST MODEL ON THE MARKET. THE SAME PROMPT TOOK FABLE 5.1 ALMOST 5 HRS AND IT CONSUMED A FRACTION OF THE TOKENS FABLE 5.1 DID.

MY FIRST GAME OUTPUT WITH GPT-6 ASTRA IT ONE-SHOTTED THIS CALL OF DUTY GAME IN 30 MINUTES!!! GPT-6 ASTRA IS THE FASTEST MODEL ON THE MARKET. THE SAME PROMPT TOOK FABLE 5.1 ALMOST 5 HRS AND IT CONSUMED A FRACTION OF THE TOKENS FABLE 5.1 DID.

hiraeth

36,888 views • 14 days ago

(1/5) Gemini 3, our most intelligent model, is landing in Google Search today – starting with AI Mode. Excited that this is the first time we’re shipping a new Gemini model in Search on day one! 🚀 In Search, Gemini 3 with generative layouts will make it easy to get a rich understanding of anything on your mind. It has state-of-the-art reasoning, deep multimodal understanding and advanced agentic capabilities. That allows the model to shine when you ask it to explain advanced concepts or ideas – it reasons and can code interactive visuals in real-time. It can tackle your toughest questions like advanced science.

(1/5) Gemini 3, our most intelligent model, is landing in Google Search today – starting with AI Mode. Excited that this is the first time we’re shipping a new Gemini model in Search on day one! 🚀 In Search, Gemini 3 with generative layouts will make it easy to get a rich understanding of anything on your mind. It has state-of-the-art reasoning, deep multimodal understanding and advanced agentic capabilities. That allows the model to shine when you ask it to explain advanced concepts or ideas – it reasons and can code interactive visuals in real-time. It can tackle your toughest questions like advanced science.

Robby Stein

94,962 views • 10 months ago

Here we go: Gemini 3.0 is live! Thankfully, I was able to be an early tester. It's the best model I've ever tested - nothing comes even close to it. OpenAI will have a very hard time to keep up with Google, because Gemini 3.0 will now be the benchmark for all future models. I have been invited by Google DeepMind / Google - Thanks to Google DeepMind for the #EarlyAccess Here's a summary of Gemini 3.0 with all the important benchmark evals:

Here we go: Gemini 3.0 is live! Thankfully, I was able to be an early tester. It's the best model I've ever tested - nothing comes even close to it. OpenAI will have a very hard time to keep up with Google, because Gemini 3.0 will now be the benchmark for all future models. I have been invited by Google DeepMind / Google - Thanks to Google DeepMind for the #EarlyAccess Here's a summary of Gemini 3.0 with all the important benchmark evals:

Chubby♨️

108,611 views • 10 months ago

This is crazy man... one prompt + 4 minutes and I hit my 5 hour rate limit in the Gemini app. Here is the proof. Yesterday, I hit the rate limit the same way, so I wanted to check if it would happen again. I started with 0% usage on my 5 hour limit, then gave one simple prompt for video generation using the avatar feature. It ran for around 3 to 4 minutes, hit 100% of the rate limit, and the video generation failed as well.

AshutoshShrivastava

82,527 views • 3 months ago

Anthropic's Head of Claude Code Boris Cherny: "Claude Code went from zero to $400 million in revenue in just a few months. it started as a hackathon project. it wasn't even meant to be a product" it still only uses the public Anthropic API. same models everyone else uses. "dog fooding successful" "over time I would actually expect cost to increase. the only thing we care about is capability. if we have to compromise capability, we won't do it" sub agents came from a Reddit post. someone built an army of agents PM, designer, frontend, backend. an engineer saw it and built it during a hackathon. it took off like wildfire internally 30 minutes. free. at Bessemer Venture Partners bookmark & watch ↓

Anthropic's Head of Claude Code Boris Cherny: "Claude Code went from zero to $400 million in revenue in just a few months. it started as a hackathon project. it wasn't even meant to be a product" it still only uses the public Anthropic API. same models everyone else uses. "dog fooding successful" "over time I would actually expect cost to increase. the only thing we care about is capability. if we have to compromise capability, we won't do it" sub agents came from a Reddit post. someone built an army of agents PM, designer, frontend, backend. an engineer saw it and built it during a hackathon. it took off like wildfire internally 30 minutes. free. at Bessemer Venture Partners bookmark & watch ↓

Carnage

162,446 views • 3 months ago