Reading · Week In Morpheus

Grok and Deepseek launches are covered in the Meta section, but there’s been so much action this week even with lesser known models.

Sabbir Hossain@Sabbirbyte

Excited to see Liquid AI drop LFM2.5-VL-3B today — a lean 3B vision-language model that reads screens, docs, charts, and the real world while running efficiently even on phones. Going to put it through its paces tonight. Looking forward to testing the grounding, tool calling,

Liquid AI @liquidai

Today, we release LFM2.5-VL-3B, a lightweight vision-language model that reads screens, documents, and the physical world. It handles digital screens across mobile, web, and desktop, grounds objects to coordinates, reads text and charts, and calls tools from either text or image

3:22 PM · Aug 12, 2026 · 272 Views

11 Likes

Suno has a new open source competitor.

rob - comfyui@hellorob

Big day for open source. MiniMax Music 3 is undeniably a state of the art open weights Music Generation model and a real alternative to Suno. My favorite part about open weights: the best is yet to come. Once the community starts playing with the model and training LoRAs, this

MiniMax (official) @MiniMax_AI

🎵MiniMax-Music3 Next-Generation Open-Weights Production-Ready & Versatile Music Model https://t.co/V2rxZk4xvh

8:10 PM · Aug 13, 2026 · 174K Views

65 Replies · 162 Reposts · 2.13K Likes

US is allowing some crazy stuff. I’m generally a fan of competition with government agencies (imagine if we could make the DMV less miserable), but you have to wonder how much of your data gets mixed up in that surveillance.

Chris Wysopal@WeldPond

This is a pretty big shift in U.S. cyber policy. The White House is setting up a program that would let private cybersecurity companies conduct government-authorized operations against foreign cybercriminal groups, including surveillance and disruption of their infrastructure.

1:29 AM · Aug 13, 2026 · 200K Views

124 Replies · 525 Reposts · 2.45K Likes

Do your piracy in peace (if you’ve got Apple, you can drop it into photos and scrub the watermark in two seconds anyway)

Guillaume Meyer@guillaumemeyer

🧹watermarks-remover now supports watermarks from OpenAI and Gemini in addition to Claude. github.com/guillaumemeyer…

github.com

GitHub - guillaumemeyer/watermarks-remover: Agent skill to strip Claude/AI text watermarks (Unicode + rewrite) and C2PA/file metadata from PNG/JPEG

8:31 PM · Aug 11, 2026 · 2.19M Views

150 Replies · 1.24K Reposts · 11.1K Likes

And last but definitely least is Smit’s question, which should intrigue anybody who’s been envying Tony Stark’s set up.

We have quite the roundup this week.

First up is our Secret Network interview. If you use AI at work, your best idea is already sitting on someone else's server. In 2023 Samsung engineers pasted proprietary chip source code into ChatGPT. In January 2026 a federal judge ordered OpenAI to hand over 20 million ChatGPT conversation logs in discovery. HSBC's answer was to ban AI internally outright — which raises the other problem: how do you compete if your people can't use the most powerful tool ever built?

Lisa Loud has spent her career on both sides of that line — Apple, PayPal, BitMEX, ShapeShift, co-founding FLUIDEFI, and now running the Secret Network Foundation. She's back on /brainpower ten months after her last appearance, and almost everything she predicted about privacy last time has already happened — faster than she expected.

She and David Johnston get specific about what actually protects an idea: confidential computing in plain language, why "you can't hand over what you don't have" is the only defense that has ever survived a subpoena, why you should never let an AI model audit its own code, what a $4.7M exploit says about AI-assisted attacks on software everyone thought was safe, and what your data is really worth to the companies collecting it.

If you're building something you can't afford to leak, start at the top.

Morpheus@MorpheusAIs

As promised. The video you have all been waiting for. @BowTiedBlueFin Gives a step by step walk through showing how easy it is to become a Morpheus Inference Provider using the Railway Install System.

5:58 PM · Aug 13, 2026 · 547 Views

1 Reply · 4 Reposts · 7 Likes

Then we have BowtiedBlueFin walking people through the Railway Install System for Morpheus inference providers. Basically, an easy way to provide compute through the Morpheus network.

Did we say buddy bots? Yes, yes we did.

Deepseek V4 Flash rocked the AI world last week, and its older brother is following in its footsteps. Not that it’s hard, with Opus 5 hitting walls like a blind racehorse.

OpenRouter@OpenRouter

DeepSeek V4 Pro 0813 is live on OpenRouter. @deepseek_ai reports large agent gains over V4 Pro Preview: DeepSWE 62.7 (+49.9), CyberGym 83.3 (+30.6), NL2Repo 61.5 (+23.0), and Terminal Bench 2.1 87.9 (+15.8) More providers coming online soon Use it now:

openrouter.ai

DeepSeek V4 Pro 0813 - API Pricing & Providers

4:38 PM · Aug 12, 2026 · 109K Views

35 Replies · 118 Reposts · 1.4K Likes

The statistics are speak for themselves. Open weight models are crushing it. Possibly because

prayag sonar@prayag_sonar

Anthropic usage dropping on OpenRouter while open-weight traffic keeps climbing feels less like “temporary routing issue” and more like people voting with their wallets after too many restriction + quality complaints. Prove me wrong with actual numbers or your own switch story.

4:41 AM · Aug 9, 2026 · 32.6K Views

42 Replies · 25 Reposts · 623 Likes

Anthropic is feeling the pressure.

Bryle@brylblzzrd

@cline

1:23 PM · Aug 13, 2026 · 8 Views

But what I didn’t expect this morning was a new Grok launch. While OpenAI spends four months between model drops, it’s been five weeks since the last Grok model. Not to mention Grok bot.