Weekly Roundup
Perle Labs Updates
Welcome to The AI Loop #22. The $PRL claim portal closed this week, Bitget futures went live, and Perle Labs enters its next chapter. Meanwhile, Meta debuted its first model from a rebuilt AI lab, AnthropicAI committed $100 million to secure the world's critical software, and Microsoft shipped three proprietary AI models built entirely without OpenAI.
Let's get into it.
Perle Labs Update
The $PRL claim portal at claim.perle.xyz officially closed on April 8 at 7am EST, marking the end of the two-week claim window. Perle Foundation ran a countdown through the week: 3 days left, final call, and a closing confirmation post that pulled 15,400 views. Thank you to everyone who completed their claim.
On April 5, Perle Foundation announced that $PRL is now live on Bitget futures, expanding leveraged trading access alongside existing perpetuals on Binance and Bybit_Official. With spot listings across Coinbase, Gate, MEXC_Official, LBank Exchange, PancakeSwap, and Binance Wallet, the token infrastructure is fully deployed. Now the focus shifts to what comes next.
Meta Debuts Muse Spark | First Model from Its Rebuilt Superintelligence Lab
Biggest quarter ever for AI, but who is getting funded?
Meta just shipped the first evidence that its AI rebuild is working. Muse Spark, released April 8, is the inaugural model from Meta Superintelligence Labs, led by former Scale AI CEO Alexandr Wang. It's a multimodal reasoning model with a "Contemplating" mode that spins up multiple parallel agents to work on different parts of a task simultaneously.
On GPQA Diamond (PhD-level reasoning), Muse Spark scored 89.5%, competitive with Claude Opus 4.6 (92.7%) and GPT-5.4 (92.8%). On HealthBench Hard, it beat every rival model at 42.8%. Meta says it achieves the same capabilities with "over an order of magnitude less compute" than Llama 4 Maverick.
Unlike Meta's previous open-weight Llama models, Muse Spark is proprietary for now, powering the Meta AI app and rolling out to WhatsApp, Instagram, Facebook, and Ray-Ban glasses. The message: Meta isn't catching up. It rebuilt from scratch and is starting a new race.
Anthropic AI Launches Project Glasswing | $100M to Secure the World's Critical Software
AnthropicAI just assembled the most powerful defensive cybersecurity coalition in tech history. Project Glasswing, announced April 8, gives partners access to Claude Mythos Preview, an unreleased frontier model, to find and fix vulnerabilities in critical software. The partner list reads like a who's who of tech infrastructure: Apple, Amazon, Google, Microsoft, NVIDIA, CrowdStrike, Cisco, Broadcom, JPMorgan, the Linux Foundation, and Palo Alto Networks, alongside ~40 more organizations.
Anthropic is committing up to $100 million in usage credits and $4 million to the Linux Foundation and Open Source Security Foundation. The model has already found "thousands of high-severity vulnerabilities, including across all major operating systems and web browsers," autonomously. Claude Mythos Preview is too powerful to release publicly; Glasswing is how Anthropic is channeling that power defensively. This is AI protecting the software the world runs on.
Source: The Verge
Microsoft Ships 3 Proprietary AI Models — Its First Built Without OpenAI
Microsoft just declared AI independence. On April 5-6, the company released MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2, its first proprietary multimodal foundation models built entirely in-house by the MAI Super Intelligence team under CEO Mustafa Suleyman. These are not wrappers around OpenAI models. They are Microsoft's own.
MAI-Transcribe-1 ranks #1 on the FLEURS benchmark with ~3.9% word error rate, beating GPT-Transcribe, Gemini 3.1 Flash, and Whisper Large v3 across 25 languages. It processes audio 2.5x faster at 50% lower cost ($0.36/hour). MAI-Voice-1 generates 60 seconds of audio in under 1 second on a single GPU with 700+ preset voices. MAI-Image-2 ranks #3 on Arena.ai, behind only Gemini 3.1 Flash and GPT Image 1.5, and is 2x faster than its predecessor. Microsoft has plans to ship a competing general-purpose LLM by 2027. The OpenAI partnership continues, but the message is clear: Microsoft is building its own stack.
Source: Windows Forum
Stay in the Loop
That's a wrap on Edition #22. Meta rebuilt its AI lab and shipped the proof. Anthropic assembled a $100M defensive coalition. Microsoft declared independence from OpenAI with its first in-house models. Back home, the $PRL claim portal closed, Bitget futures are live, and the next chapter begins.
Join the @PerleLabs community on X, Discord, and Telegram to stay connected and catch the next drop of insights from across the ecosystem.