Sensitive content

This media may contain sensitive content.

正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Hello Gemini 2.5 Flash-Lite! So fast, it codes each screen on the fly (Neural OS concept 👇). The frontier isn't always about large models and beating benchmarks. In this case, a super fast & good model can unlock drastic use cases. Read more:

Oriol Vinyals

180,428 subscribers

405,746 次观看 • 1 年前 •via X (Twitter)

科学技术教育

Anya Rossi• Live Now

Private livecam show

10 条评论

Serçiya^ سەرچیا 的头像

Serçiya^ سەرچیا1 年前

What are some useful cases for this?

AJ - e/acc 🚀🇨🇴🇵🇷 的头像

AJ - e/acc 🚀🇨🇴🇵🇷1 年前

It's FAST!

HarrisonAIX 的头像

HarrisonAIX1 年前

what's the Gemini VScode extension using nowadays? Gemini 2.5 Flash-Lite? From what I gather, users can't justify ultra and pro is a necessity. The extension being updated would be next level.

Chris 的头像

Chris1 年前

Apps could just be APIs, where the users can just ask for whatever UI they want and the LLMs will build it

Suman 的头像

Suman1 年前

While aiming for better responsiveness with Neural OS ideas, we need to think carefully about how well the model works in different situations, keep user data safe in complex systems, and ensure clear rules for user experiences that change.

Chanchal 💬 的头像

Chanchal 💬1 年前

mind blowing 😲

BlockRadar 的头像

BlockRadar1 年前

Alpha tech, WAGMI

Jesahn O 的头像

Jesahn O1 年前

Best use cases?

Djasnive Rajaona 的头像

Djasnive Rajaona1 年前

@GoogleDeepMind What the heck

Sic Semper Salustri🏴‍☠️ 的头像

Sic Semper Salustri🏴‍☠️1 年前

@GoogleDeepMind 👀

相关视频

Introducing a brand new way to use Jan 💛 You can now search the web, do deep research, and let Jan use your browser. Powered by Jan-v2-VL-Max, 30B model beating Gemini 2.5 Pro & DeepSeek R1 on execution benchmarks. Try it:

👋 Jan

29,403 次观看 • 7 个月前

Free is the part people are underpricing. Google just made Gemini 3.6 Flash the new default. Not a toy model hiding in a free tier. A near-frontier coding model normal users can try in Gemini and AI Studio without touching enterprise pricing. This is landing while people compare paid frontier APIs: > GPT-5.6 Sol at $5/$30 > Claude Sonnet 5 at $3/$15 > Grok 4.5 at $2/$6 > Gemini 3.5 Flash at $1.50/$9 Benchmarks matter. But a model people can actually use for free changes behavior faster than a model they only admire from pricing tables.

Free is the part people are underpricing. Google just made Gemini 3.6 Flash the new default. Not a toy model hiding in a free tier. A near-frontier coding model normal users can try in Gemini and AI Studio without touching enterprise pricing. This is landing while people compare paid frontier APIs: > GPT-5.6 Sol at $5/$30 > Claude Sonnet 5 at $3/$15 > Grok 4.5 at $2/$6 > Gemini 3.5 Flash at $1.50/$9 Benchmarks matter. But a model people can actually use for free changes behavior faster than a model they only admire from pricing tables.

ＺＥＦＩ

13,074 次观看 • 16 天前

Fast enough to play games ⚡️ We challenged Gemini 3 Flash to a quick drawing game in the G3mini. Because this model is optimized for speed, it can see and guess what you’re drawing while you’re still sketching it.

Fast enough to play games ⚡️ We challenged Gemini 3 Flash to a quick drawing game in the G3mini. Because this model is optimized for speed, it can see and guess what you’re drawing while you’re still sketching it.

Google

119,117 次观看 • 7 个月前

Playing Pokemon with LLMs became a benchmark, but what about Super Mario? Turns out Google DeepMind Gemini 2.0 Flash can play Super Mario in real-time due to its fast latency, multimodal input and long context! 🎮 Gemini 2.0 Flash receives screenshots of the game and then generates Python code (PyAutoGUI commands) for each screenshot to control the game for a short period (the next 1 or 2 seconds). Add a while loop to it and you have it play in real-time. 🤯

Playing Pokemon with LLMs became a benchmark, but what about Super Mario? Turns out Google DeepMind Gemini 2.0 Flash can play Super Mario in real-time due to its fast latency, multimodal input and long context! 🎮 Gemini 2.0 Flash receives screenshots of the game and then generates Python code (PyAutoGUI commands) for each screenshot to control the game for a short period (the next 1 or 2 seconds). Add a while loop to it and you have it play in real-time. 🤯

Philipp Schmid

153,086 次观看 • 1 年前

🚨 Gemini 3.5 Flash checkpoint notes: The old “cool Gemini UI” vibe feels mostly gone. It is more GPTified now. Good: fast, smart for a Flash model. Bad: it keeps filling websites with HUDs/panels, even when the prompt asks for restraint. Biggest issue: prompt adherence. I told it multiple times not to use web, and it still did. Because of that, exact knowledge cutoff is hard to verify confidently.

🚨 Gemini 3.5 Flash checkpoint notes: The old “cool Gemini UI” vibe feels mostly gone. It is more GPTified now. Good: fast, smart for a Flash model. Bad: it keeps filling websites with HUDs/panels, even when the prompt asks for restraint. Biggest issue: prompt adherence. I told it multiple times not to use web, and it still did. Because of that, exact knowledge cutoff is hard to verify confidently.

Chetaslua

127,743 次观看 • 2 个月前

Who said LLMs can only generate text? I gave 5 frontier models the ABC notation of Canon in D (the fast part) and asked each to play it in their own style. > GPT 5.6, Fable 5 & Opus 5 extended it and wrote their own endings > Gemini 3.6 Flash kept the notes, swapped the instruments > Kimi K3 rewrote the whole thing My pick: Fable's version sounds surprisingly rich. Opus's version is great too. Sound on 🔊

Who said LLMs can only generate text? I gave 5 frontier models the ABC notation of Canon in D (the fast part) and asked each to play it in their own style. > GPT 5.6, Fable 5 & Opus 5 extended it and wrote their own endings > Gemini 3.6 Flash kept the notes, swapped the instruments > Kimi K3 rewrote the whole thing My pick: Fable's version sounds surprisingly rich. Opus's version is great too. Sound on 🔊

stevibe

11,921 次观看 • 9 天前

Love this example! Congrats to Google DeepMind! I successfully replicated the Boggle game on SWE Arena. TLDR: Gemini 2.0 Pro won! In the case of Claude 3.5 Sonnet vs Gemini 2.0 Pro, both are given the image and prompt to generate the algo to get the results. The image and prompt are exactly the same. Try it out on SWE Arena: BTW can we get a bigger streaming rate limit for our API, so that ppl can better compare the latest Gemini models with other frontier models? Jeff Dean Logan Kilpatrick 👀

Love this example! Congrats to Google DeepMind! I successfully replicated the Boggle game on SWE Arena. TLDR: Gemini 2.0 Pro won! In the case of Claude 3.5 Sonnet vs Gemini 2.0 Pro, both are given the image and prompt to generate the algo to get the results. The image and prompt are exactly the same. Try it out on SWE Arena: BTW can we get a bigger streaming rate limit for our API, so that ppl can better compare the latest Gemini models with other frontier models? Jeff Dean Logan Kilpatrick 👀

Terry Yue Zhuo

48,220 次观看 • 1 年前

Loading DeepSeek v4 Flash just reminded me how good and fast this model is and how well it runs on DGX Sparks. GLM-5.2 is really awesome, but if you want fast and concurrent inference, DS4F is still a top choice IMO. Created by DS4F in single shot The coffee guy

Loading DeepSeek v4 Flash just reminded me how good and fast this model is and how well it runs on DGX Sparks. GLM-5.2 is really awesome, but if you want fast and concurrent inference, DS4F is still a top choice IMO. Created by DS4F in single shot The coffee guy

Mia

39,686 次观看 • 10 天前

✨Spotlight: Top 4 Code Arena open models Open models are moving fast, and this community is directly shaping how that progress shows up in real use. Your builds, votes, and experiments surface what frontier models can actually do in practice. To celebrate the momentum, we’re spotlighting the top open models in Code Arena (WebDev) 🧵👇 🔥 MiniMax (official) — MiniMax-M2.1 🤖 Z.ai — GLM-4.7 ✨ DeepSeek — Deepseek-v3.2-thinking 🚀 Xiaomi MiMo — Mimo-v2-flash (non-thinking) Let’s keep building in the open and raising the bar together 🚀

✨Spotlight: Top 4 Code Arena open models Open models are moving fast, and this community is directly shaping how that progress shows up in real use. Your builds, votes, and experiments surface what frontier models can actually do in practice. To celebrate the momentum, we’re spotlighting the top open models in Code Arena (WebDev) 🧵👇 🔥 MiniMax (official) — MiniMax-M2.1 🤖 Z.ai — GLM-4.7 ✨ DeepSeek — Deepseek-v3.2-thinking 🚀 Xiaomi MiMo — Mimo-v2-flash (non-thinking) Let’s keep building in the open and raising the bar together 🚀

Arena.ai

16,742 次观看 • 7 个月前

I just had to make a new video of GLM-4.7-Flash - Helping me refactor VLLM studio - Did a data analytics report for work - Managed to search my tweets - Made me a fully playable Pacman in 1 shot - Great at browser use This model is too good to be this small, the full thing will fit on a macbook, it's fast, precise and can do pretty much anything Sonnet can. It somehow benches higher than Sonnet 4, 3.7, Opus 4.5 no thinking, all the GPT models from last year and more.

I just had to make a new video of GLM-4.7-Flash - Helping me refactor VLLM studio - Did a data analytics report for work - Managed to search my tweets - Made me a fully playable Pacman in 1 shot - Great at browser use This model is too good to be this small, the full thing will fit on a macbook, it's fast, precise and can do pretty much anything Sonnet can. It somehow benches higher than Sonnet 4, 3.7, Opus 4.5 no thinking, all the GPT models from last year and more.

0xSero

38,263 次观看 • 6 个月前

Sundar Pichai just dropped a surprisingly candid look into Google DeepMind’s current AI heartbeat. Here’s the essence of the clip: • Google is now operating on a steady six month cadence, pushing the frontier twice a year • Sundar openly admits that meaningful leaps are getting harder • Gemini 2.5 Pro was already a beast, so improving on it is no small task • Developers love the Flash models because they scale efficiently for millions of users • Sundar is genuinely excited about Gemini 3.0 Flash and even hints it might be their best model yet • While the public focuses on current releases, internal teams are already pre training the next generation • This culture of relentless innovation is pushing the entire stack toward major breakthroughs heading into 2026 It really feels like we’re watching the early stages of a world where AI takes over more and more of the work so humans can focus on… well, living.

Sundar Pichai just dropped a surprisingly candid look into Google DeepMind’s current AI heartbeat. Here’s the essence of the clip: • Google is now operating on a steady six month cadence, pushing the frontier twice a year • Sundar openly admits that meaningful leaps are getting harder • Gemini 2.5 Pro was already a beast, so improving on it is no small task • Developers love the Flash models because they scale efficiently for millions of users • Sundar is genuinely excited about Gemini 3.0 Flash and even hints it might be their best model yet • While the public focuses on current releases, internal teams are already pre training the next generation • This culture of relentless innovation is pushing the entire stack toward major breakthroughs heading into 2026 It really feels like we’re watching the early stages of a world where AI takes over more and more of the work so humans can focus on… well, living.

VraserX e/acc

162,001 次观看 • 8 个月前

🚨 New Paper Hybrid Neural World Models. We build neural world models for discontinuous environments and our method discovers when they're making error, with no ground truth. Our neural models run physics fast, leaping many simulator steps in a single pass. How we find points where not to trust it: ask the model for the same future two ways, and the disagreement between the two answers marks where it's failing without needing labels. The same recipe holds on three systems with nothing in common: a chemical wave, compressible gas with shocks, rigid bodies colliding in 3D. 👇

🚨 New Paper Hybrid Neural World Models. We build neural world models for discontinuous environments and our method discovers when they're making error, with no ground truth. Our neural models run physics fast, leaping many simulator steps in a single pass. How we find points where not to trust it: ask the model for the same future two ways, and the disagreement between the two answers marks where it's failing without needing labels. The same recipe holds on three systems with nothing in common: a chemical wave, compressible gas with shocks, rigid bodies colliding in 3D. 👇

Lossfunk

15,037 次观看 • 2 个月前

This week, we tested 3 latest models in our Game Arena Benchmark: → O3 → O4-mini → Gemini 2.5 Flash Across 4 games—Phoenix Wright, Sokoban, Candy Crush, and 2048—O3 dominated the zero-shot leaderboard, ranking #1 or #2 in nearly every task and outperforming previous SOTA models like O3-mini and Gemini 2.5 Flash. 🔥 Beyond our customized tests, we took on a real challenge: Sokoban (1989)—the classic, unforgiving original. 🗣️ Many say O3 shows strong image reasoning, but does it really? Let’s see what our Game Arena Benchmark reveals. 👇

This week, we tested 3 latest models in our Game Arena Benchmark: → O3 → O4-mini → Gemini 2.5 Flash Across 4 games—Phoenix Wright, Sokoban, Candy Crush, and 2048—O3 dominated the zero-shot leaderboard, ranking #1 or #2 in nearly every task and outperforming previous SOTA models like O3-mini and Gemini 2.5 Flash. 🔥 Beyond our customized tests, we took on a real challenge: Sokoban (1989)—the classic, unforgiving original. 🗣️ Many say O3 shows strong image reasoning, but does it really? Let’s see what our Game Arena Benchmark reveals. 👇

Hao AI Lab

15,105 次观看 • 1 年前

Libghostty can now be used to fuzz TUIs, thanks to Oskar Wickström + Antithesis. They already found bugs in multiple including btop. I always imagined libghostty would be useful for testing TUIs, super happy to see this is both practical and valuable. This is another example of where speed matters! "Why does Ghostty need to be so fast?" Well, if you're running hundreds or thousands of unit tests that each use a clean in-memory terminal, you want that to be fast. If you're fuzz testing and trying to push an unlimited amount of data through a terminal, you want that terminal to be fast. So many people got hung up on "why does my terminal _GUI_ need to be fast" without connecting one more dot and realizing the GUI is only fast if the core is fast, and the core being fast unlocks a hell of a lot more. Like this.

Libghostty can now be used to fuzz TUIs, thanks to Oskar Wickström + Antithesis. They already found bugs in multiple including btop. I always imagined libghostty would be useful for testing TUIs, super happy to see this is both practical and valuable. This is another example of where speed matters! "Why does Ghostty need to be so fast?" Well, if you're running hundreds or thousands of unit tests that each use a clean in-memory terminal, you want that to be fast. If you're fuzz testing and trying to push an unlimited amount of data through a terminal, you want that terminal to be fast. So many people got hung up on "why does my terminal _GUI_ need to be fast" without connecting one more dot and realizing the GUI is only fast if the core is fast, and the core being fast unlocks a hell of a lot more. Like this.

Mitchell Hashimoto

71,584 次观看 • 3 个月前

THE AI RACE IS OVER. DUST CLEARED. GROK 4 FAST ALREADY HIT 67 BILLION TOKENS Grok 4 Fast just bulldozed the leaderboard, clocking 67.1 billion tokens and leaving rivals gasping for bandwidth! Leaderboard Mode: • Qwen3’s 80B trails at 23B – miles behind • Gemini 2.5 Flash barely flickers in 3rd • Claude? Still waiting at the loading screen • TrustAI charts show Grok’s surge lighting up November It’s not just an AI model anymore – it’s a speed-run on the entire industry! Source: TrustAI, xAI, Medium

THE AI RACE IS OVER. DUST CLEARED. GROK 4 FAST ALREADY HIT 67 BILLION TOKENS Grok 4 Fast just bulldozed the leaderboard, clocking 67.1 billion tokens and leaving rivals gasping for bandwidth! Leaderboard Mode: • Qwen3’s 80B trails at 23B – miles behind • Gemini 2.5 Flash barely flickers in 3rd • Claude? Still waiting at the loading screen • TrustAI charts show Grok’s surge lighting up November It’s not just an AI model anymore – it’s a speed-run on the entire industry! Source: TrustAI, xAI, Medium

Mario Nawfal

196,729 次观看 • 8 个月前

How to use 50+ API keys (models) for FREE on OpenClaw API??? - go to - login or register your account - click on "more models" - click on "use case" and select what you need it for - choose the model and open it - click on "view code" → "Generate API key" many models don't allow direct deploy, so use the "view code" button to generate API access basically Nvidia NIM gives you the ability to test almost any model from their list for FREE some of them are not worse than GPT 5.2 or Claude Opus 4.6, some might even perform better depending on the task how to understand if a model is efficient and compare it with others??? - go to - type the model name in search - click on "benchmarks" - you’ll see performance tests and rankings this way you can easily compare free models with paid ones of course there are RPM limits, on many models it’s around ~40 requests per minute each model is different, after generating the API key, RPM limits are shown in the top-right corner nothing stops you from using them, many work perfectly fine, super solid option for first tests and for learning OpenClaw or any other system where you need an AI API model

How to use 50+ API keys (models) for FREE on OpenClaw API??? - go to - login or register your account - click on "more models" - click on "use case" and select what you need it for - choose the model and open it - click on "view code" → "Generate API key" many models don't allow direct deploy, so use the "view code" button to generate API access basically Nvidia NIM gives you the ability to test almost any model from their list for FREE some of them are not worse than GPT 5.2 or Claude Opus 4.6, some might even perform better depending on the task how to understand if a model is efficient and compare it with others??? - go to - type the model name in search - click on "benchmarks" - you’ll see performance tests and rankings this way you can easily compare free models with paid ones of course there are RPM limits, on many models it’s around ~40 requests per minute each model is different, after generating the API key, RPM limits are shown in the top-right corner nothing stops you from using them, many work perfectly fine, super solid option for first tests and for learning OpenClaw or any other system where you need an AI API model

Ronin

58,846 次观看 • 5 个月前

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

kwindla

40,593 次观看 • 2 个月前

You can now try Llama 3.1 405B for free (link below)! This is the largest open-source model out there, and for the first time, an open model is competitive with closed models. This time around, Meta did something new: Llama 3.1 has a license that allows developers to use it to enhance other models. For the first time, you can distill Llama 3.1 405B's capabilities into a smaller, more practical model for your use case. First, here is the link where you can play with Llama 3.1 for free: The model is hosted in Tune Studio, an end-to-end platform for developing applications using Large Language Models. They are sponsoring this post. Take a look at the attached video. It will show you how you can fine-tune a simple model using Llama 3.1 without leaving the platform: 1. You can create an empty dataset 2. Use the playground to generate and record interactions with Llama 3.1 3. Modify the dataset directly using the playground 4. Export the data and fine-tune a smaller model Fast and easy! As long as you have a web browser, you can start experimenting with fine-tuning and Llama 3.1. That's all it takes!

You can now try Llama 3.1 405B for free (link below)! This is the largest open-source model out there, and for the first time, an open model is competitive with closed models. This time around, Meta did something new: Llama 3.1 has a license that allows developers to use it to enhance other models. For the first time, you can distill Llama 3.1 405B's capabilities into a smaller, more practical model for your use case. First, here is the link where you can play with Llama 3.1 for free: The model is hosted in Tune Studio, an end-to-end platform for developing applications using Large Language Models. They are sponsoring this post. Take a look at the attached video. It will show you how you can fine-tune a simple model using Llama 3.1 without leaving the platform: 1. You can create an empty dataset 2. Use the playground to generate and record interactions with Llama 3.1 3. Modify the dataset directly using the playground 4. Export the data and fine-tune a smaller model Fast and easy! As long as you have a web browser, you can start experimenting with fine-tuning and Llama 3.1. That's all it takes!

Santiago

55,609 次观看 • 2 年前