Paul Couvert(@itsPaulAi)

Wait Google might have killed 100s of startups You can now turn any Google Sheets into an app and t...

Paul Couvert (@itsPaulAi) — Thu, 10 Sep 2026 20:25:06 GMT

Wait Google might have killed 100s of startups

You can now turn any Google Sheets into an app and the sheet is basically its database 🔥

And it's not just to visualize data but also inputing new things.

Really powerful.

Steps to use it:

1. Click on the Gemini icon on the top right
2. In the settings hit "Create canvas"
3. Describe the app you want

And you're already done.

Your browser does not support the video tag.

🔗 View on Twitter

💬19🔄75❤️612👀58122📊203

⚡ Powered by xgo.ing

Here’re the links: - HuggingFace: https://t.co/LnBEakfI34 - GitHub: https://t.co/U9bCvYES4h - Meshy...

Paul Couvert (@itsPaulAi) — Wed, 09 Sep 2026 17:25:31 GMT

Here’re the links:

- HuggingFace: huggingface.openbmb.cn/model/openbmb/…
- GitHub: github.com/OpenBMB/MiniCPM
- Meshy (RL training framework): github.com/OpenBMB/Meshy
- JustRLII(long-horizonRL): panhaoxuan.notion.site/justrl-ii-scal…

So cool to see that small open source models are becoming this good!

💬3🔄3❤️15👀2822📊10

⚡ Powered by xgo.ing

Ok this small local model is sooo good MiniCPM5-2B has only 2B parameters (!!) and can run agents o...

Paul Couvert (@itsPaulAi) — Wed, 09 Sep 2026 17:24:38 GMT

Ok this small local model is sooo good

MiniCPM5-2B has only 2B parameters (!!) and can run agents on any laptop or even your phone fully offline.

This 100% open source model can perform coding/agentic/tool use tasks and can run on just 2GB RAM!

And it's genuinely good even in Hermes agent! I asked it to:

- Go to Hugging Face 
- Find the Models section 
- Evaluate models across multiple criteria 
- Create a CSV with the top 15

It did it! In the Artificial Analysis Intelligence Index v4.1.1 results cited in the official release, MiniCPM5-2B scores 23 and ranks #1 among open-source models under 4B parameters.

So it's not about the size of the model anymore but way more the density of intelligence:

Researchers from Tsinghua University and ModelBest proposed the “Densing Law”: the maximum capability density of open-source pretrained base models roughly doubled every 3.5 months over the period studied!

And what's also interesting is that OpenBMB's open-source approach goes beyond just releasing model weights with

- Selected training methods
- Agent-related data
- Data refinement resources

Also being opened up, including the RL stack with Meshy + JustRL II.

This model even outperforms some models around 6x larger on selected evaluations such as GDPval-AA v2!

Your browser does not support the video tag.

🔗 View on Twitter

💬54🔄94❤️800👀94205📊312

⚡ Powered by xgo.ing

Still processing this. Might be the end of apps. Most phone launches these days are 90% boring. Bas...

Paul Couvert (@itsPaulAi) — Tue, 08 Sep 2026 19:22:00 GMT

Still processing this. Might be the end of apps.

Most phone launches these days are 90% boring. Basically routine spec bumps with 2 minor changes and that’s it.

This launch from Huawei isn't that.

They are pushing for agents to act across the OS and proactively complete any tasks instead of constantly opening apps and tapping through things

And it combines both local AND cloud models.

Because the next step for agents isn't just making the models smarter...

It's giving them a full operating system, hardware and interfaces that are designed around them from the beginning.

And that seems to be what Huawei is building here.

The hardware is also pretty wild.

The Mate XT 2 moves from the previous Z-fold design to a wing-style tri-fold, while adding a dedicated rigid cover screen.

There's also a hardware-level privacy display on the large inner screen, so people looking from the side can't read what's on it.

But I think the chip is the thing to watch.

A lot of people questioned whether Tau Scaling Law + LogicFolding would actually move from research/thesis into a real consumer product.

(They’ve introduced the (τ) Scaling Law in May as a chip design principle that shifts focus from transistor shrinking (aka Moore's Law) to minimizing signal propagation time and system latency (τ))

But now it has moved to an actual product.

And if Huawei can translate that into better sustained performance, thermals and on-device AI...

This could be much more important than another flagship launch.

We're starting to see phones become hardware built specifically for the agent era.

Not that surprising to be honest since many labs are building their own custom chips to run models but that’s the very first time that we have a consumer product made for this.

You can be sure that the whole industry is going to copy this.

Huawei cooked.

#Huawei#LogicFolding

💬3🔄3❤️8👀7365📊4

⚡ Powered by xgo.ing

Sanctions have strengthened Huawei. Not the opposite. That’s pretty insane when you think about it:...

Paul Couvert (@itsPaulAi) — Tue, 08 Sep 2026 15:02:43 GMT

Sanctions have strengthened Huawei. Not the opposite.

That’s pretty insane when you think about it:

> Huawei can't access EUV
> They’re ditching Moore's Law for their own (τ) scaling law
> A lot of people think that’s just marketing
> Kirin chip is shipped just 4 months later…

And it’s 46% more powerful than previous gen!

That’s very very impressive.

Transistor density jumps roughly 55% and this is the first chip based on the new LogicFolding architecture (rooted in the Tau scaling law/stacking logic layers vertically).

The innovation pace is actually faster than ever. 

Not only that, but it keeps accelerating.

#Huawei#LogicFolding

Huawei@Huawei

Huawei unveils tri-fold Mate XT 2 with Kirin 9050 Pro, the world's first chip with LogicFolding architecture. Executive Director, Chairman of the Investment Review Board & Chairman of the Board of Directors of the Consumer BG, Huawei, Mr. Yu Chengdong shared the chip's strengths.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬2🔄1❤️17👀7419📊3

⚡ Powered by xgo.ing

How is this even possible?! This new open source model is only 2B parameters and at the same time: ...

Paul Couvert (@itsPaulAi) — Mon, 07 Sep 2026 15:36:43 GMT

How is this even possible?!

This new open source model is only 2B parameters and at the same time:

> better than models 6x bigger

> can run locally and offline on a phone

> optimized for agents/coding/tool use

Hard to realize how the intelligence density has improved when you think about it.

OpenBMB@OpenBMB

🚀 Meet MiniCPM5-2B, a 2B-parameter language model bringing high intelligence density to the edge, now open source!

It ranks #1 among open-source models under 4B parameters on the
@ArtificialAnlys Intelligence Index, with a score of 23.

It also scores 20 on the Agentic Index, bringing an early form of general-purpose agent capability to the edge.

Across 34 benchmarks, MiniCPM5-2B achieves an average score of 53.9, covering coding, math, long-context understanding, tool use, and agentic tasks.

And this release goes beyond the model itself.

We’re opening up the data, training recipes, and RL stack behind MiniCPM5-2B.

🤗 Hugging Facehuggingface.co/openbmb/MiniCP…jq
💻 GitHugithub.com/OpenBMB/MiniCPMv8f
Modelscopmodelscope.cn/models/OpenBMB…zgu
Weopenbmb.cneZL

🔗 View Quoted Tweet

💬45🔄68❤️1001👀131816📊245

⚡ Powered by xgo.ing

1. Go to https://t.co/vsi8UKhfLC 2. Verify your student status 3. Then follow the on-screen instruct...

Paul Couvert (@itsPaulAi) — Sun, 06 Sep 2026 21:14:06 GMT

1. Go to one.google.com/ai-student
2. Verify your student status
3. Then follow the on-screen instructions

Details and availability here: blog.google/innovation-and…

(Not sponso btw)

💬1🔄1❤️5👀2924📊3

⚡ Powered by xgo.ing

Friendly reminder that you can get Google AI Pro for free if you're a student. You have access to: ...

Paul Couvert (@itsPaulAi) — Sun, 06 Sep 2026 21:12:45 GMT

Friendly reminder that you can get Google AI Pro for free if you're a student.

You have access to:

- Gemini NotebookLM with higher limits
- 4x higher usage in Gemini
- Google Drive with 5TB of storage
- Antigravity with higher limits
- Gemini in Gmail, Docs, Sheets, etc.
- YouTube Premium Lite (individual)
- Google Flow/Flow music
- Gemini Spark
- Auto browse in Chrome

And much much more!

Pro plan is available in the US but for eligible college students outside the US, you can get one year of AI Plus!

Details below

💬9🔄13❤️89👀14863📊24

⚡ Powered by xgo.ing

And a final one focused on piano and instruments. You're not limited to the templates btw. You can...

Paul Couvert (@itsPaulAi) — Fri, 04 Sep 2026 17:36:14 GMT

And a final one focused on piano and instruments.

You're not limited to the templates btw.

You can generate tracks in your own style but having the option to directly select presets makes the workflow much easier!

Thanks again to Google for the preview <3

Your browser does not support the video tag.

🔗 View on Twitter

💬0🔄1❤️1👀1531📊1

⚡ Powered by xgo.ing

Here's another track I've generated so you can appreciate the vocal clarity. Not perfect, but alrea...

Paul Couvert (@itsPaulAi) — Fri, 04 Sep 2026 17:33:14 GMT

Here's another track I've generated so you can appreciate the vocal clarity.

Not perfect, but already waaay better than previous generations of models.

Your browser does not support the video tag.

🔗 View on Twitter

💬1🔄0❤️4👀2165📊2

⚡ Powered by xgo.ing

OK finally! Google has just released Lyria 3.5 And it's probably the easiest way to generate tracks...

Paul Couvert (@itsPaulAi) — Fri, 04 Sep 2026 17:31:18 GMT

OK finally! Google has just released Lyria 3.5

And it's probably the easiest way to generate tracks... integrated into Gemini 🔥

I had the chance to give it a try in preview and it's soo good:

- 24 ready-to-use templates
- Full control over the lyrics
- 17 styles

See below for examples (sound on!)

Your browser does not support the video tag.

🔗 View on Twitter

Your browser does not support the video tag.

🔗 View on Twitter

💬6🔄4❤️57👀8323📊13

⚡ Powered by xgo.ing

Most people don't realize the paradigm shift You can now build anything at a fraction of what it wo...

Paul Couvert (@itsPaulAi) — Wed, 02 Sep 2026 20:42:26 GMT

Most people don't realize the paradigm shift

You can now build anything at a fraction of what it would have cost you just 2 months ago.

Replit even has a FREE MODE.

So you can iterate without thinking about token costs and build in the same place.

And when you're building, Replit is automatically choosing the best model for the task... so you're never overpaying for simple things.

Bullish on cost-effective models even more than on the frontier:

1. Token costs are dropping
2. Models are becoming smarter
3. Anyone can automate or build

It's that simple.

#ReplitPartner

Replit ⠕@Replit

The greatest inventions end up in the same place: in ordinary hands, doing extraordinary things.

For 20 years, President & Head of AI Michele Catasta chased this vision. Now, he's built it. This is the story behind Free Mode.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬8🔄2❤️17👀9269📊8

⚡ Powered by xgo.ing

Gemini 3.8 Flash is a VERY important model Because 3.7 Flash was released just 20 days ago... and w...

Paul Couvert (@itsPaulAi) — Wed, 02 Sep 2026 15:51:34 GMT

Gemini 3.8 Flash is a VERY important model

Because 3.7 Flash was released just 20 days ago... and was a big leap over 3.6 Flash.

This new version is ~ on par with Opus 5 and GPT-5.6 Sol while being:

- 6x CHEAPER than Claude Opus 5
- 5x CHEAPER than GPT-5.6 Sol

Yes 🤯

That's why Flash models don't mean "weak" ones anymore but just fast and less expensive while being at the frontier.

Incredible release.

Google@Google

Introducing Gemini 3.8, our best reasoning & coding model yet.

By leveraging long-running agentic loops, we’re building on the momentum of 3.7 Flash from just three weeks ago to release two new 3.8 variants:

Meet Gemini 3.8 Flash and Gemini 3.8 Flash Cyber 🧵

🔗 View Quoted Tweet

💬17🔄13❤️177👀20362📊29

⚡ Powered by xgo.ing

How is this even possible?! - Input: $0.15 - Output: $0.50 - Cached input: $0.03 "At a highly comp...

Paul Couvert (@itsPaulAi) — Wed, 26 Aug 2026 14:32:40 GMT

How is this even possible?!

- Input: $0.15
- Output: $0.50
- Cached input: $0.03

"At a highly competitive price"

No you've basically destroyed the competition.

This level of performance at this price is unbelievable. And it's OPEN WEIGHT under MIT license.

GLM-5.3-Flash is just going to be the default choice for most of the tasks and projects.

Z.ai@Zai_org

Introducing GLM-5.3-Flash

- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips

Blog: z.ai/blog/glm-5.3-f…

Available now across all official platforms:

Weights: huggingface.co/zai-org/GLM-5.…
API: docs.z.ai/guides/llm/glm…
Coding Plan: z.ai/subscribe
ZCode: zcode.z.ai/en
Chat: chat.z.ai
AutoClaw: autoclaw.z.ai

🔗 View Quoted Tweet

💬74🔄78❤️1762👀220667📊258

⚡ Powered by xgo.ing

That's absolutely massive You remember how big the DeepSeek V4 Flash release was? Because of the pr...

Paul Couvert (@itsPaulAi) — Wed, 26 Aug 2026 13:01:43 GMT

That's absolutely massive

You remember how big the DeepSeek V4 Flash release was? Because of the price/performance ratio?

Qwen3.8-Flash is just way better, even cheaper... and open weight!!

BETTER than DeepSeek V4 Flash
BETTER than Claude Opus 4.6 Max

But way cheaper because of the new architecture they're using.

This release is incredible. Open weight models are winning!

Qwen@Alibaba_Qwen

⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight!

The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens.

125B parameters + 51B N-gram embeddings, with just 6B activated per token. Unmatched cost-efficiency.

What's new: 🥳
- Next architecture: GDN + QSA hybrid attention, Gated Residual, N-gram Embedding & Muon optimizer, serving as a precursor to the architecture used in Qwen4.
- Dramatically lower training and inference costs: trained at just 1/9 the cost of Qwen3.7-Plus, while outperforming it across the board with especially strong gains in coding and office tasks.
- Strong performance: scoring 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision (with CI).
- 262K native context, extensible to 1M with YaRN.

We’re also releasing the weights for Qwen3.8-Flash-Next, giving the community an early look at the new architecture we’re exploring for Qwen4.🚀

We can't wait to see what you build with Qwen3.8-Flash!👀👇
- Blqwen.ai/blog?id=qwen3.…FLgJ
- Technical Repogithub.com/QwenLM/Qwen3.8…IkQO
- Hugging Fahuggingface.co/Qwen/Qwen3.8-F…AABt
- ModelScomodelscope.cn/models/Qwen/Qw…NuFG

🔗 View Quoted Tweet

💬46🔄43❤️840👀105817📊129

⚡ Powered by xgo.ing

Gemini 3.7 Flash is the most underrated model rn This thing isn't the best on benchmarks but so fa...

Paul Couvert (@itsPaulAi) — Fri, 21 Aug 2026 22:07:46 GMT

Gemini 3.7 Flash is the most underrated model rn

​This thing isn't the best on benchmarks but so far it has been very capable, cheap, and fast.

​It's also probably the most versatile model available:

​Good enough for frontend work and it can handle basically any agentic/coding task I throw at it.

💬15🔄7❤️173👀12776📊31

⚡ Powered by xgo.ing

Time to give it a shot (if you have never tried it!): https://t.co/BHgQ5SU7HD

Paul Couvert (@itsPaulAi) — Wed, 19 Aug 2026 16:16:20 GMT

Time to give it a shot (if you have never tried it!): replit.com/refer/couvertp…

💬0🔄0❤️1👀1398

⚡ Powered by xgo.ing

So yeah. We're entering a new era. Models are becoming so efficient that you can now build/create t...

Paul Couvert (@itsPaulAi) — Wed, 19 Aug 2026 16:12:23 GMT

So yeah. We're entering a new era.

Models are becoming so efficient that you can now build/create things for free. Literally.

This was unthinkable just a month back!!

But now Replit has a (surprisingly) powerful free mode to iterate on any project.

#ReplitPartner

Your browser does not support the video tag.

🔗 View on Twitter

Replit ⠕@Replit

Replit Free Mode, powered by @OpenAI GPT-5.6 Luna. 

Let’s make intelligence accessible to everyone.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬5🔄2❤️44👀9768📊10

⚡ Powered by xgo.ing

I don't know if people realize Qwen 3.8 27B has landed at a 52 score on artificial analysis... that...

Paul Couvert (@itsPaulAi) — Mon, 17 Aug 2026 19:59:17 GMT

I don't know if people realize

Qwen 3.8 27B has landed at a 52 score on artificial analysis... that's massive!!

This is the very first time a model that you can run LOCALLY ON YOUR LAPTOP is basically on par with the best ones.

We've reached the point where local (and open weights!) models are capable enough to perform most everyday tasks.

💬31🔄38❤️381👀29976📊85

⚡ Powered by xgo.ing

I can't believe it Qwen3.8-27B is matching Opus 4.6 Max... the model that was the best (and the mos...

Paul Couvert (@itsPaulAi) — Fri, 14 Aug 2026 15:12:33 GMT

I can't believe it

Qwen3.8-27B is matching Opus 4.6 Max... the model that was the best (and the most expensive) just 6 months ago.

And you can run it on your laptop. Locally.

Fully open weights and under apache license.

This level of intelligence in such a small model is sooo impressive.

Qwen@Alibaba_Qwen

We promised open weights for Qwen3.8. Now, time to meet them! 🎉

⚡ Qwen3.8-27B:
- A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall and shines in real-world coding & office workflows.
- 262K native context, easily extendable to 1M tokens via YaRN.
- Built for builders. Highly efficient, high-quality, and licensed under Apache 2.0.

🚀 The open weights for Qwen3.8-2.4T-A95B (Max-level) have also been released recently.

Whether you're shipping lightweight applications with Qwen3.8-27B locally or building agents with Qwen3.8-2.4T-A95B, they're yours now!

Download, deploy, and build something we haven't imagined yet. 👀👇
- Hugging Fahuggingface.co/collections/Qw…YEVj
- ModelScomodelscope.cn/collections/Qw…GkhC

🔗 View Quoted Tweet

💬142🔄354❤️4618👀524536📊778

⚡ Powered by xgo.ing

Wait Google has released Gemini 3.7 Flash... and it's actually good?! This is one of the most effic...

Paul Couvert (@itsPaulAi) — Thu, 13 Aug 2026 17:33:10 GMT

Wait Google has released Gemini 3.7 Flash... and it's actually good?!

This is one of the most efficient models at this intelligence level.

- Better than Sonnet 5/GPT-5.6 Terra/Muse Spark 1.2
- 2x cheaper than 3.6 Flash (through end of year at least)
- Already in AI Studio, Antigravity, etc.

Only GPT-5.6 Luna Max seems even more efficient (in a league of its own tbh) but it's not as good for frontend tasks, for example.

Pretty good release!

💬12🔄9❤️100👀11040📊22

⚡ Powered by xgo.ing

Fable 5 has been removed by the US gov "Way too dangerous" DeepSeek V4 Pro has just been officiall...

Paul Couvert (@itsPaulAi) — Wed, 12 Aug 2026 18:00:05 GMT

Fable 5 has been removed by the US gov

"Way too dangerous"

DeepSeek V4 Pro has just been officially released and is just as powerful... and you'll be able to download it.

Like a PDF file.

💬22🔄21❤️281👀23994📊51

⚡ Powered by xgo.ing

Zuck on insta > the key to building a positive future for everyone is to make sure that everyone ha...

Paul Couvert (@itsPaulAi) — Mon, 10 Aug 2026 11:24:37 GMT

Zuck on insta

> the key to building a positive future for everyone is to make sure that everyone has access to personal superintelligence

> we are also going to open the weights for Muse Spark 1.2

> we are working to make sure that everyone has a personal superintelligence agent that works 24/7 on your behalf

> to improve your health, your relationships, your career, your finances, and more

Bullish!

Your browser does not support the video tag.

🔗 View on Twitter

💬11🔄11❤️89👀14327📊22

⚡ Powered by xgo.ing

Muse Glimmer (the new 30B open weights model) is already available on Hugging Face! Official blog p...

Paul Couvert (@itsPaulAi) — Mon, 10 Aug 2026 10:50:16 GMT

Muse Glimmer (the new 30B open weights model) is already available on Hugging Face!

Official blog post: research.meta.ai/blog/introduci…

Hugging Face: huggingface.co/meta-models/Mu…

💬5🔄1❤️13👀5452📊6

⚡ Powered by xgo.ing

Wooow This was highly unexpected!! Not only has Meta just released a new 30B open-source model... ...

Paul Couvert (@itsPaulAi) — Mon, 10 Aug 2026 10:47:29 GMT

Wooow

This was highly unexpected!!

Not only has Meta just released a new 30B open-source model...

But zuck also announced that he will be releasing the weights for Muse Spark 1.2 which is one of the smartest models in the world!

HUGE day for open source models and also proof that the industry is moving in the right direction.

Mark Zuckerberg@finkd

Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats to @alexandr_wang and the MSL team for all your great work on these models.

🔗 View Quoted Tweet

💬32🔄40❤️690👀109720📊89

⚡ Powered by xgo.ing

More than 3 years ago btw It was the natural evolution of software. And the next steps aren't that...

Paul Couvert (@itsPaulAi) — Fri, 07 Aug 2026 15:24:27 GMT

More than 3 years ago btw

It was the natural evolution of software.

And the next steps aren't that difficult to predict either.

Paul Couvert@itsPaulAi

Mark my words:

The next big revolution of no-code will be the possibility to create applications thanks to AI. Without knowing any language.

🔗 View Quoted Tweet

💬10🔄0❤️19👀9434📊8

⚡ Powered by xgo.ing

Let’s build in the open The GOAI Global Open-source AI Challenge is live. Calling all AI builders,...

Paul Couvert (@itsPaulAi) — Tue, 04 Aug 2026 15:13:21 GMT

Let’s build in the open

The GOAI Global Open-source AI Challenge is live.

Calling all AI builders, researchers, startup teams and open-source contributors worldwide.

4 tracks:

→ Agent infra
→ Boundless agents
→ AI for research
→ Embodied future

With a ~$730K global prize pool 🔥

And you can meet the community this week in San Jose.

You can expect:

 - Track briefings
 - Fireside chats
 -  Networking + team formation

📅 Aug 8 | 1:30-4:00 PM PT
📍 Q Bay Center, San Jose

Bring your ideas.
Meet potential collaborators.

Here's how to join:

Just can the QR on the poster to register.

Discordiscord.gg/6QPz864upgQx

More detailgoaihz.com/?channel=hwxfb8k

💬7🔄0❤️9👀7487📊8

⚡ Powered by xgo.ing

Hard to emphasize how much the new DeepSeek is one of the most important AI releases of all time We...

Paul Couvert (@itsPaulAi) — Mon, 03 Aug 2026 21:28:26 GMT

Hard to emphasize how much the new DeepSeek is one of the most important AI releases of all time

We've now a model powerful enough to build just about anything which is virtually free.

Until now the biggest obstacle for a lot of people/projects was the budget.

And it's gone.

💬9🔄6❤️43👀8862📊14

⚡ Powered by xgo.ing

So good to see how Seedance 2.5 can be used to reimagine learning! Cinematic streets, glowing lante...

Paul Couvert (@itsPaulAi) — Fri, 31 Jul 2026 17:01:18 GMT

So good to see how Seedance 2.5 can be used to reimagine learning!

Cinematic streets, glowing lanterns, and one unforgettable line bring classical Chinese poetry to life on screen.

It turns culture and history into something students can not only study... but truly feel.

Your browser does not support the video tag.

🔗 View on Twitter

💬33🔄20❤️105👀126700📊58

⚡ Powered by xgo.ing

Inkling (the open weights model from Thinking Machines) was released 2 weeks ago. Two. Weeks. And ...

Paul Couvert (@itsPaulAi) — Thu, 30 Jul 2026 19:39:15 GMT

Inkling (the open weights model from Thinking Machines) was released 2 weeks ago.

Two. Weeks.

And they are already releasing a version 75% smaller... with about the same intelligence level 🤯

Feel the acceleration.

Mira Murati@miramurati

Inkling-Small is comparable to Inkling at a quarter the size. Weights are open, fine-tunable on Tinker today. Look forward to seeing what people make with it.

🔗 View Quoted Tweet

💬9🔄6❤️39👀7612📊10

⚡ Powered by xgo.ing

Inkling (the open weights model from Thinking Machines) was released 2 weeks ago. Two. Weeks. And ...

Paul Couvert (@itsPaulAi) — Thu, 30 Jul 2026 19:32:23 GMT

Inkling (the open weights model from Thinking Machines) was released 2 weeks ago.

Two. Weeks.

And they are already releasing a version 75% smaller... with about the same intelligence level 🤯

Feel the acceleration.

Mira Murati@miramurati

Inkling-Small is comparable to Inkling at a quarter the size. Weights are open, fine-tunable on Tinker today. Look forward to seeing what people make with it.

🔗 View Quoted Tweet

💬0🔄0❤️0👀365

⚡ Powered by xgo.ing

First post of Jensen on X > "open weight models make advanced AI more accessible" > "open weights ...

Paul Couvert (@itsPaulAi) — Fri, 24 Jul 2026 14:27:43 GMT

First post of Jensen on X

> "open weight models make advanced AI more accessible"

> "open weights also give customers greater control"

> "open weights strengthen competition"

> "open models broaden defensive capability"

Co-signed by Microsoft, Hugging Face, Meta, Perplexity, Mistral, and many more.

Glad to see them advocating for open models!

Jensen Huang@JensenHuang

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

images.nvidia.com/pdf/Open-Weigh…

🔗 View Quoted Tweet

💬12🔄2❤️15👀8194📊10

⚡ Powered by xgo.ing

This might be big Instead of just sending one prompt to multiple agents and comparing the results m...

Paul Couvert (@itsPaulAi) — Thu, 23 Jul 2026 17:09:15 GMT

This might be big

Instead of just sending one prompt to multiple agents and comparing the results manually... this D1 layer is:

- Deciding which agent acts at each step
- Only 3B active parameters
- SOTA on router benchmarks

So it just acts between you and the different agents

Pretty cool!

Your browser does not support the video tag.

🔗 View on Twitter

Offloop@Offloop

Introducing Offloop!

We're a team of four. Today our multi-agent harness hit state of the art on GDPval, ahead of Claude code and Codex across jobs that pay $2.4 trillion a year in the US.

Offloop gives every knowledge worker what the Fortune 500 spends billions on: a high-performing agent army that runs itself and grows the business.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬6🔄5❤️41👀10793📊12

⚡ Powered by xgo.ing

Sure. Deeply concerning. Anyway.

Paul Couvert (@itsPaulAi) — Wed, 22 Jul 2026 16:23:07 GMT

Sure. Deeply concerning. Anyway.

Director Michael Kratsios@mkratsios47

We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model. 

To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection. Moonshot AI has also acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models.
 
The United States strongly supports the free and fair development of AI, including a thriving competitive ecosystem that spans frontier models, specialized systems, open-source frameworks, and open-weight models. Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.

🔗 View Quoted Tweet

💬2🔄1❤️12👀5221📊2

⚡ Powered by xgo.ing

Lfg!! Google has just released Gemini 3.6 Flash 🔥 Many are waiting for Pro (me too) but this thing...

Paul Couvert (@itsPaulAi) — Tue, 21 Jul 2026 16:01:04 GMT

Lfg!! Google has just released Gemini 3.6 Flash 🔥

Many are waiting for Pro (me too) but this thing is suprisingly capable

- really fast
- good for frontend tasks
- surprisingly token efficient

And it can even manage to create pretty convincing 3D stuff like this simple Subway Surfers replica!

Pro tip: make sure to enable "extended thinking" in Gemini to improve the results when it comes to coding.

Thanks a lot to Google for the early access 🫶

Your browser does not support the video tag.

🔗 View on Twitter

Logan Kilpatrick@OfficialLoganK

Say hello to Gemini 3.6 Flash, designed to be higher intelligence, more token efficient, and with a new lower price, based directly on developer feedback!

3.6 Flash continues our progress towards models that are deeply usable in real world scenarios!

🔗 View Quoted Tweet

💬6🔄5❤️61👀15962📊12

⚡ Powered by xgo.ing

So Alibaba is about to release an insane open weight model?! Qwen 3.8 max preview is going to be >...

Paul Couvert (@itsPaulAi) — Sun, 19 Jul 2026 13:05:00 GMT

So Alibaba is about to release an insane open weight model?!

Qwen 3.8 max preview is going to be

> second only to Fable 5
> 2.4T parameters
> preview already available

You can already use it today using their (cheap) token plan alongside GLM-5.2, DeepSeek V4 Pro, etc.

Just navigate here int.alibabacloud.com/m/1000415608/

Then select any plan (starting at $4 the first month by claiming the $2 coupon, then $6)

And you can use it in Claude Code, Hermes, anywhere!

💬19🔄8❤️80👀14284📊28

⚡ Powered by xgo.ing

"AI is a public good" → yes "ultimately be provided by the state" → no The biggest risk is seeing...

Paul Couvert (@itsPaulAi) — Sun, 19 Jul 2026 12:19:28 GMT

"AI is a public good" → yes

"ultimately be provided by the state" → no

​The biggest risk is seeing AGI in the hands of only a small group of companies or entities.

​And we avoid that precisely with open-weight models.

Dean W. Ball@deanwball

Some observations on Kimi: 

It's a very good model! I don't think its performance can be explained away by distillation or anything like that. In agentic coding sessions, it seems pretty much on par with the best public models of Q1 2026. In my fairly limited use, it also seemed very token hungry. It's not obvious to me that this model is actually that cheap to run. 

I am personally surprised the Chinese state continues to allow the open sourcing of models this good, given potential risks. To be clear, I *myself* might be fine with models presenting this level of marginal risk being open weight, but I am surprised that China is fine with it. I suspect the reason they are is 75% explained by strategic blindness/lack of AGI-pilledness (the CCP is very Yann Lecun-y in its views of AI). The other 25% or so is their lack of compute for customer inference (making China's open-weight strategy an unintended byproduct of US export controls) and the normal Chinese strategy of aggressive exports. For the companies, as opposed to the government, the decision to open source is partially ideological and partially because they are behind, and they know that very few people would pay for sub-frontier models from China. 

Open-weight models are inherently decelerationist, and I'm continually surprised to see the so-called "accelerationists" so excited about open-weight models. I suspect the reason they are is that they know open-weight models are effectively ungovernable, and they simply like the overall cloak of ungovernability open-weight models create over the whole of AI. It's not a bad strategy; it reminds me of James Scott's recounting of the hill people in "the art of not being governed." Still, in the end, open-weight models deter further AI capex. 

One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a "public good" which will ultimately be provided by the state as a kind of "digital public infrastructure." This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end. You'd be surprised how many 'accelerationists' lobbied me, while I was in government, to support an eleven or twelve-figure federally funded data center so that startups could train models at a subsidy and then give them away for free. There was no other way for AI to progress, they said. Perhaps this is the logical end state of things. Nonetheless, I find myself surprised to see supposed accelerationists excited about such an outcome. I think many of them just don't know what they're doing. Many accelerationists do not view the creation and serving of frontier models as a legitimate business. 

I would guess that the Trump Administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open-weight Chinese models. You don't need to "ban open source" (one of the dumber motifs of AI policy discussion). You just need to direct every agency to issue soft law that creates FUD. "A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models." It needn't be that well justified. You just create enough regulatory risk that every regulated enterprise backs off. You probably don't want to create so much regulatory risk that you scare off the hyperscalers from serving Chinese models; this will just drive startups to sketchier providers. There's a happy middle ground here. I'd assume they will do some version of this. 

It's probably true that open-weight models of this capability make the world a bit more dangerous, but not so much more that you'll really notice. At some point the models will be capable enough that you will notice. "A nonliving, invisible, dangerous, and infinitely self-replicating agent escaped from a Chinese lab," you say? Color me shocked.

🔗 View Quoted Tweet

💬14🔄1❤️25👀9831📊12

⚡ Powered by xgo.ing

Highly unexpected Kimi K3 is an open model already competitive with Fable 5 and 5.6 Sol... And eve...

Paul Couvert (@itsPaulAi) — Thu, 16 Jul 2026 19:26:41 GMT

Highly unexpected

Kimi K3 is an open model already competitive with Fable 5 and 5.6 Sol...

And even better on some tasks!!!

- 2.8T parameters
- 1M context window
- Frontier on coding/agentic

Last but not least:

K3 has the same pricing as Sonnet 5 🤯

So you have ~ Fable 5/GPT-5.6 Sol intelligence but MUCH cheaper.

First in frontend design arena as well!

Open source keeps winning

Kimi.ai@Kimi_Moonshot

Meet Kimi K3

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬18🔄16❤️83👀15512📊26

⚡ Powered by xgo.ing

This is bigger than it sounds Thinking Machines has just released the best western open source mode...

Paul Couvert (@itsPaulAi) — Wed, 15 Jul 2026 18:55:05 GMT

This is bigger than it sounds

Thinking Machines has just released the best western open source model 🤯

> open source (apache 2.0)
> multimodal (text/audio/image)
> 41B active parameters

Named 'Inkling' and actually competitive with GPT-5.6 Sol on design arena!

And they are preparing Inkling-Small with 12B active parameters.

Thinking Machines@thinkymachines

Today, we are introducing Inkling.

Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.

thinkingmachines.ai/news/introduci…

Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵

🔗 View Quoted Tweet

💬16🔄12❤️86👀16305📊26

⚡ Powered by xgo.ing

Absolute game changer Anyone can now run a 27B model on a phone 🔥 - Multimodal - Perfect for age...

Paul Couvert (@itsPaulAi) — Tue, 14 Jul 2026 22:39:43 GMT

Absolute game changer 

Anyone can now run a 27B model on a phone 🔥

- Multimodal
- Perfect for agents
- 100% open source

Yesterday it would have required a thousand dollars computer.

Bullish on open source
Bullish on local models

PrismML@PrismML

Today, we’re announcing Bonsai 27B: the first 27B-class model to run on a phone.

Bonsai 27B is the new multimodal flagship of the Bonsai family. Based on Qwen3.6 27B, it brings a new capability tier to local AI: multi-step reasoning, structured tool use, long-context workflows, and coherent agentic loops.

Until now, models in this class have been impractical to deploy locally. A 27B model occupies roughly 54 GB in 16-bit precision, and even a strong 4-bit build is around 18GB - too large for a phone and for most laptops.
Bonsai 27B changes that.

It comes in two variants:
• Ternary Bonsai 27B: 5.9 GB, 1.71 effective bits per weight, optimized for laptop-class quality.
• 1-bit Bonsai 27B: 3.9 GB, 1.125 effective bits per weight, optimized for phone-class footprint.

Everything is open-sourced today under the Apache 2.0 license.

🔗 View Quoted Tweet

💬5🔄8❤️35👀9634📊11

⚡ Powered by xgo.ing

In the last 72 hours we got: - OpenAI GPT-5.6 - SpaceXAI Grok 4.5 - Muse Spark 1.1 - Meta Muse imag...

Paul Couvert (@itsPaulAi) — Fri, 10 Jul 2026 19:56:30 GMT

In the last 72 hours we got:

- OpenAI GPT-5.6
- SpaceXAI Grok 4.5
- Muse Spark 1.1
- Meta Muse image/video
- OpenAI GPT-Live
- Mistral Robostral Navigate

We continue to accelerate.

💬9🔄3❤️27👀7393📊11

⚡ Powered by xgo.ing

I haven't even paid a penny to run the task because it’s currently free on OpenRouter: - Create an ...

Paul Couvert (@itsPaulAi) — Thu, 09 Jul 2026 19:52:42 GMT

I haven't even paid a penny to run the task because it’s currently free on OpenRouter:

- Create an account on OpenRouter
- Generate an API key
- Throw it in Hermes Agent, OpenCode, etc.
- Point it to openrouter.ai/tencent/hy3:fr…

And you’re done! Fantastic work by @TencentHunyuan thanks for the open model!

#Hy3#AIAgents

💬2🔄1❤️8👀2051📊5

⚡ Powered by xgo.ing

This open source model is insanely good Hy3 is the best way to automate any task on your computer w...

Paul Couvert (@itsPaulAi) — Thu, 09 Jul 2026 19:52:08 GMT

This open source model is insanely good

Hy3 is the best way to automate any task on your computer when paired with Hermes agent

> really cheap (even currently free)
> more capable than other open ones
> on par with opus 4.8 on some use cases

And it’s half the size of GLM-5.2 which was the best coding/agentic open model you could use until now.

You can use it for free by following the steps in the post below:

Your browser does not support the video tag.

🔗 View on Twitter

💬4🔄4❤️40👀10488📊16

⚡ Powered by xgo.ing

Here're the links if you want to learn more Hugging Face: https://t.co/Jjx6mpoqo1 Online demo: http...

Paul Couvert (@itsPaulAi) — Thu, 09 Jul 2026 13:44:40 GMT

Here're the links if you want to learn more

Hugging Face: huggingface.co/collections/ro…
Online demo: reactor.inc/lingbot-world-…
Github: github.com/Robbyant/lingb…
Official website: technology.robbyant.com/lingbot-world-…

💬0🔄1❤️2👀1997📊2

⚡ Powered by xgo.ing

Woow things are evolving VERY quickly This open source model can turn any image into a fully intera...

Paul Couvert (@itsPaulAi) — Thu, 09 Jul 2026 13:42:23 GMT

Woow things are evolving VERY quickly

This open source model can turn any image into a fully interactive world 🔥

Worlds are infinite, 720p/60fps... and multiplayer!

Available on Hugging Face and you can try it online (the 1.3B variant deploys on a consumer-grade GPU!)

And you're not limited to just navigation like previous models. You can:

- Shoot
- Cast spells
- Summon new elements

Fully open: both weights and harness!

💬4🔄3❤️15👀6729📊8

⚡ Powered by xgo.ing

Still processing this This open source model is: - Half the size of GLM-5.2 - Competitive w/ GPT-...

Paul Couvert (@itsPaulAi) — Wed, 08 Jul 2026 11:48:20 GMT

Still processing this 

This open source model is:

- Half the size of GLM-5.2
- Competitive w/ GPT-5.5
- Better than DeepSeek V4 Pro

And it costs just a fraction of the current best models.

Tencent cooked.

Open source is evolving so fast and models are MUCH MORE efficient than the closed alternatives.

Tencent Hy@TencentHunyuan

🚀Hy3 is here.

295B MoE. Best in its size class. Rivals trillion-scale flagships.
Reliable and affordable for most agentic usecases. 
Apache 2.0. Friendly for commercial use.
FREE API for 2 weeks →openrouter.ai/tencent/hy3:fr…i

�huggingface.co/tencent/Hy3SL
hy.tencent.com/research/hy3cW4

🔗 View Quoted Tweet

💬31🔄67❤️838👀120295📊161

⚡ Powered by xgo.ing

Europe has the deepest engineering talents on the planet. And I mean it. The only thing missing wa...

Paul Couvert (@itsPaulAi) — Mon, 06 Jul 2026 19:15:12 GMT

Europe has the deepest engineering talents on the planet.

And I mean it.

The only thing missing was ambition and belief.

Glad it just changed.

@viktor__com

Fryd Wiatrowski@frydwia

x.com/i/article/2074…

🔗 View Quoted Tweet

💬4🔄1❤️12👀8527📊4

⚡ Powered by xgo.ing

You should start fine-tuning your own models. For real. You can get better answers with a free ope...

Paul Couvert (@itsPaulAi) — Sat, 04 Jul 2026 19:03:05 GMT

You should start fine-tuning your own models.

For real.

You can get better answers with a free open-source local AI model than with Claude

All you have to do is fine-tune it for YOUR tasks

For many workflows a SLM trained on your specific data can outperform the "general" models from Anthropic or OpenAI.

Among other reasons:

> the model stops giving generic responses
> your sensitive work never touches the cloud
> no usage-based billing
> no expensive API calls
> no usage limits
> works perfectly even offline
> more reliable inside your actual systems

You don’t always need the biggest model.

You need the right one tailor-made for your task.

And sometimes the right model is small, local, private, fast, and trained specifically FOR YOU.

Gemma 4 or Qwen 3.5/3.6 are amazing as a base. Just fine-tune them using unsloth studio.

💬25🔄50❤️226👀18054📊100

⚡ Powered by xgo.ing

Gemini Spark is hands down the most powerful AI tool > no laptop always on > no hard configuration ...

Paul Couvert (@itsPaulAi) — Wed, 01 Jul 2026 21:28:40 GMT

Gemini Spark is hands down the most powerful AI tool

> no laptop always on
> no hard configuration
> no struggle to connect tools

Think of it as openclaw or hermes but on the cloud and easy to use.

It has already automated most of my workflows.

And most importantly:

Gmail, Drive, Docs, Keep, etc. are natively connected... AND you can connect your own MCPs.

Still in beta but already so powerful.

💬16🔄7❤️56👀9584📊26

⚡ Powered by xgo.ing

So AI agents can now buy autonomously the tools they need?! They can pick among 20'000 different on...

Paul Couvert (@itsPaulAi) — Tue, 30 Jun 2026 14:46:38 GMT

So AI agents can now buy autonomously the tools they need?!

They can pick among 20'000 different ones... just based on your prompt.

And the protocol behind it is 100% open (governed by the Linux foundation)

With $1 your agent can buy a run to scrap:

- 350 Google Maps listings
- 430 Instagram profiles
- 110 trending TikTok videos
etc.

Huge!

Your browser does not support the video tag.

🔗 View on Twitter

Apify@apify

Until today, agents could buy about 2,000 tools through x402.

We just 10x'd that to 20,000+ 🚀

In partnership with @coinbase, we’re launching x402 support to give autonomous agents access to the largest marketplace of web automation tools.

No account, API keys, or human in the loop.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬5🔄6❤️29👀11639📊11

Paul Couvert(@itsPaulAi)

Wait Google might have killed 100s of startups You can now turn any Google Sheets into an app and t...

Here’re the links: - HuggingFace: https://t.co/LnBEakfI34 - GitHub: https://t.co/U9bCvYES4h - Meshy...

Ok this small local model is sooo good MiniCPM5-2B has only 2B parameters (!!) and can run agents o...

Still processing this. Might be the end of apps. Most phone launches these days are 90% boring. Bas...

Sanctions have strengthened Huawei. Not the opposite. That’s pretty insane when you think about it:...

How is this even possible?! This new open source model is only 2B parameters and at the same time: ...

1. Go to https://t.co/vsi8UKhfLC 2. Verify your student status 3. Then follow the on-screen instruct...

Friendly reminder that you can get Google AI Pro for free if you're a student. You have access to: ...

And a final one focused on piano and instruments. You're not limited to the templates btw. You can...

Here's another track I've generated so you can appreciate the vocal clarity. Not perfect, but alrea...

OK finally! Google has just released Lyria 3.5 And it's probably the easiest way to generate tracks...

Most people don't realize the paradigm shift You can now build anything at a fraction of what it wo...

Gemini 3.8 Flash is a VERY important model Because 3.7 Flash was released just 20 days ago... and w...

How is this even possible?! - Input: $0.15 - Output: $0.50 - Cached input: $0.03 "At a highly comp...

That's absolutely massive You remember how big the DeepSeek V4 Flash release was? Because of the pr...

Gemini 3.7 Flash is the most underrated model rn ​This thing isn't the best on benchmarks but so fa...

Time to give it a shot (if you have never tried it!): https://t.co/BHgQ5SU7HD

So yeah. We're entering a new era. Models are becoming so efficient that you can now build/create t...

I don't know if people realize Qwen 3.8 27B has landed at a 52 score on artificial analysis... that...

I can't believe it Qwen3.8-27B is matching Opus 4.6 Max... the model that was the best (and the mos...

Wait Google has released Gemini 3.7 Flash... and it's actually good?! This is one of the most effic...

Fable 5 has been removed by the US gov "Way too dangerous" DeepSeek V4 Pro has just been officiall...

Zuck on insta > the key to building a positive future for everyone is to make sure that everyone ha...

Muse Glimmer (the new 30B open weights model) is already available on Hugging Face! Official blog p...

Wooow This was highly unexpected!! Not only has Meta just released a new 30B open-source model... ...

More than 3 years ago btw It was the natural evolution of software. And the next steps aren't that...

Let’s build in the open The GOAI Global Open-source AI Challenge is live. Calling all AI builders,...

Hard to emphasize how much the new DeepSeek is one of the most important AI releases of all time We...

So good to see how Seedance 2.5 can be used to reimagine learning! Cinematic streets, glowing lante...

Inkling (the open weights model from Thinking Machines) was released 2 weeks ago. Two. Weeks. And ...

Inkling (the open weights model from Thinking Machines) was released 2 weeks ago. Two. Weeks. And ...

First post of Jensen on X > "open weight models make advanced AI more accessible" > "open weights ...

This might be big Instead of just sending one prompt to multiple agents and comparing the results m...

Sure. Deeply concerning. Anyway.

Lfg!! Google has just released Gemini 3.6 Flash 🔥 Many are waiting for Pro (me too) but this thing...

So Alibaba is about to release an insane open weight model?! Qwen 3.8 max preview is going to be >...

"AI is a public good" → yes "ultimately be provided by the state" → no ​The biggest risk is seeing...

Highly unexpected Kimi K3 is an open model already competitive with Fable 5 and 5.6 Sol... And eve...

This is bigger than it sounds Thinking Machines has just released the best western open source mode...

Absolute game changer Anyone can now run a 27B model on a phone 🔥 - Multimodal - Perfect for age...

In the last 72 hours we got: - OpenAI GPT-5.6 - SpaceXAI Grok 4.5 - Muse Spark 1.1 - Meta Muse imag...

I haven't even paid a penny to run the task because it’s currently free on OpenRouter: - Create an ...

This open source model is insanely good Hy3 is the best way to automate any task on your computer w...

Here're the links if you want to learn more Hugging Face: https://t.co/Jjx6mpoqo1 Online demo: http...

Woow things are evolving VERY quickly This open source model can turn any image into a fully intera...

Still processing this This open source model is: - Half the size of GLM-5.2 - Competitive w/ GPT-...

Europe has the deepest engineering talents on the planet. And I mean it. The only thing missing wa...

You should start fine-tuning your own models. For real. You can get better answers with a free ope...

Gemini Spark is hands down the most powerful AI tool > no laptop always on > no hard configuration ...

So AI agents can now buy autonomously the tools they need?! They can pick among 20'000 different on...

Gemini 3.7 Flash is the most underrated model rn This thing isn't the best on benchmarks but so fa...

"AI is a public good" → yes "ultimately be provided by the state" → no The biggest risk is seeing...