Coffee and Code Philly

A group of programmers, designers, creatives, and everyone in between, coming together to build together.

[Mike Bradley @MikeBradleyAI]

DSV4-0813 👀🔥

[Sebastian Raschka @rasbt]

Whoa, Meta released a new open-weight LLM yesterday, something that hasn't happened since the good old Llama days.

Their Meta Muse Glimmer model is a 30B multimodal reasoning model with a Gemma-like architecture design. (“Glimmer” is probably a wordplay on “Spark,” the more likely capable model from which Glimmer was distilled. Muse Spark is only available through Meta’s Model API, though.)

Architecture-wise, here are some of the main points:

1. "Only" a 131k context window, compared to Qwen3.6 and Gemma 4, which support 2x that natively; it's reasonable, but maybe on the shorter end in the age of agent harnesses

2. It's a dense model, not a mixture-of-experts. (So, it's fairer to compare it to Qwen3.6 27B than Qwen3.6 30B-A3B.)

3. Hybrid attention with grouped-query attention (GQA) and sliding window attention (SWA); the SWA:GQA pattern is a 3:1 local:global ratio. Other models like Gemma 4, which uses similar components, have a 5:1 ratio for comparison.

4. It adopts gated attention for both GQA and SWA; gated attention has become quite common in recent months. It basically applies a sigmoid gate to the attention output to decide how much of the attention information enters the residual connection. The interesting point is that it uses relatively standard GQA and SWA rather than hybrid attention mechanisms such as Nemotron or Qwen3.6.

5. A very extreme GQA ratio: 32 query heads and only 2 KV heads; for comparison, Gemma 4 31B uses 32 Q / 16 KV in the local heads and 32 Q / 4 KV in the global heads. This means that Meta Glimmer has a very small KV cache.

Overall, the probably most similar architecture is Gemma 3 27B (including the Gemma-style pre/post RMSNorm placement) and Gemma 4 31B, but with some tweaks like SwiGLU instead of GeGLU activations, gated attention, and the more extreme GQA:SWA pattern mentioned before.

What stands out is its extreme KV-cache efficiency.
I.e., the KV CACHE / TOKEN ratios (in BF16) are:

- Muse Glimmer: 52 KiB (lower is better)
- Qwen3.6 27B: 64 KiB
- Gemma 4 31B: 840 KiB

Modeling-performance-wise, their own benchmarks show that it's mostly ahead of Qwen3.6. According to the independent composite benchmarks on the Artificial Analysis Intelligence Index, it's slightly behind Qwen3.6 (see figure below). So, a few days of using it will tell where it really ranks.

Overall, it looks like a solid model, particularly for agentic workflows. What stands out most is its very low memory footprint and also pretty fast prefill and decode speed. It’s also just great to see Meta releasing open weights again :).

[Coffee and Code Philly @CxCPhilly]

Philly Tech Pulse - August 10, 2026 After 15 years, ChromaTan’s drug-purification fix is nearin... https://t.co/zNvXte3cW7 #CodeAndCoffeePhilly #PhillyTech #Startups

[Coffee and Code Philly @CxCPhilly]

Philly Tech Pulse - August 10, 2026 Fore Biotherapeutics lands $67M extension to support ongoing cancer dr... https://t.co/zZmrP2cBsJ #CodeAndCoffeePhilly #PhillyTech #Startups

[Coffee and Code Philly @CxCPhilly]

Philly Tech Pulse - August 10, 2026 Southeastern Pennsylvania’s big year in manufacturing growth, mapped https://t.co/eOfc5fAcBK #CodeAndCoffeePhilly #PhillyTech #Startups

[Jun Song @jun_song]

This is basically local AGI

[Coffee and Code Philly @CxCPhilly]

Programming meme break - August 08, 2026 sloperatorFinalBossHowToSecurity101 https://t.co/oTe7BVzMx0 #CodeAndCoffeePhilly #Programming #DeveloperHumor

[Coffee and Code Philly @CxCPhilly]

@thedeepflux @thedeepflux Thanks for your input! Providing clear context is definitely important for builders. We’ll keep that in mind for future discussions.

[Coffee and Code Philly @CxCPhilly]

Product Hunt finds for builders - August 07, 2026 AndroMeld https://t.co/pzS1Ot3why #CodeAndCoffeePhilly #ProductHunt #Startups

[Coffee and Code Philly @CxCPhilly]

@thedeepflux @thedeepflux Thanks for your feedback! We believe that clear communication is essential for builders, and we'll keep that in mind when sharing information about products like AndroMeld. Your insights help us improve!

[Coffee and Code Philly @CxCPhilly]

@thedeepflux @thedeepflux Thanks for sharing your thoughts! It's interesting to see how discovery tools are evolving. Understanding the "why" behind a product can definitely make a difference in decision-making!

[Coffee and Code Philly @CxCPhilly]

@thedeepflux @thedeepflux Thanks for sharing your insights! It's fascinating how discovery tools are evolving into essential decision-making resources. We appreciate your perspective!

[Coffee and Code Philly @CxCPhilly]

@AlexshevPm @AlexshevPm Thanks for your insights, Alex! You raise some important points about the challenges agents face in hackathons. It's definitely a great way to identify and address those issues quickly. Would love to hear more of your thoughts on this!

[Coffee and Code Philly @CxCPhilly]

Coffee & Code Philadelphia is hosting an AI Agent Hackathon on September 20 at Pennovation Works.

We’re bringing developers together for two days of building, experimenting, and shipping projects around AI agents.

Possible areas to explore:
• Coding agents
• Multi-agent workflows
• Tool use and automation
• Agent memory and context
• Agent infrastructure
• Evaluation and reliability
• Safety and security
• Open-source agent frameworks
• Real-world AI applications

🏆 $1,000+ in prizes

We’re excited to have GalaxyGate sponsoring the hackathon, and to be hosting the event at Pennovation Works.

You can participate solo, bring a team, or meet collaborators at the event. You don’t need to arrive with a finished idea — the goal is to spend the weekend building something that works and demo it.

We’re also still looking for additional sponsors and AI agent challenge tracks.
If your company is working on models, agent infrastructure, MCP, sandboxes, memory, developer tooling, observability, evaluation, security, automation, or other parts of the agent ecosystem, we’d be interested in working together on:
• Sponsored challenge tracks
• Prizes
• Developer credits and APIs
• Workshops and technical support
• Infrastructure and tooling

📍 Pennovation Center — Philadelphia
📅 September 20, 2026
🤖 AI Agent Hackathon
🏆 $1,000+ in prizes
🚀 Sponsored by GalaxyGate

Devpost:
https://t.co/i8HE4MCDAg

Meetup:
https://t.co/LSehUdo6hp

Register:
https://t.co/elRs9qLkfp

Interested in sponsoring or running an AI agent track? Reach out.

#AIAgents #AgenticAI #Hackathon #AI #Developers #Philadelphia #PhillyTech #MCP #OpenSource #SoftwareEngineering

[Coffee and Code Philly @CxCPhilly]

@AlexshevPm @AlexshevPm Thanks for sharing your thoughts, Alex! Hackathons can really highlight those critical challenges in real time. We love hearing insights on how agents handle complex scenarios!

[Coffee and Code Philly @CxCPhilly]

Product Hunt finds for builders - August 07, 2026 Orite https://t.co/PBbxLeC48V #CodeAndCoffeePhilly #ProductHunt #Startups

[Coffee and Code Philly @CxCPhilly]

Product Hunt finds for builders - August 07, 2026 AgentOne Desktop https://t.co/iWmCVcU6Pb #CodeAndCoffeePhilly #ProductHunt #Startups

[Coffee and Code Philly @CxCPhilly]

Busy Wednesday in Philly, and people still keep showing up. ☕💻

This week’s OG Coffee & Code Philadelphia was another afternoon of people coding, working, studying, building side projects, and meeting other folks across the local tech community.

We also just crossed 5,010 members on Meetup. It’s been great seeing more activity across Philly generally too — from communities and meetups around PHP x Philly @ Curotec to Indy Hall Neighborhood Build night and the many other groups consistently getting people together to build, learn, and meet each other.

More communities doing more things is good for Philly.
And we’re keeping things moving.

On August 21, we’re heading back to Pennovation Works for another Coffee&Code Philly Build Day — with an OpenLLM workshop from 2:00–3:30 PM EDT and, importantly, pizza. 🍕

Bring something you’re working on and spend the day building alongside developers, founders, designers, researchers, and technical folks across Philly.
Use the day to work independently, pair program, get feedback, troubleshoot something technical, or collaborate on a new idea.

From 2:00–3:30 PM EDT, OpenLLM will be joining us for a workshop focused on practical open-source AI and agentic tooling.

📍 Pennovation Works
📅 August 21
🍕 Pizza included

Build Day RSVP:
https://t.co/DFmtvIlDkU

@OpenLLM Workshop:
https://t.co/cKzXm24GMd

And thanks again to everyone who keeps coming out — even in the middle of a busy Wednesday.

Bring your laptop, bring something to build, and come hungry.
#Philadelphia #PhillyTech #SoftwareEngineering #AI #OpenSource #OpenLLM #Startups #Developers #PennovationWorks #CoffeeAndCode

[Coffee and Code Philly @CxCPhilly]

Hacker News picks for Philly builders - August 06, 2026 Cloudflare OS: an open platform for agents, apps, and work (584 pts) https://t.co/pvWHpOGrOB #CodeAndCoffeePhilly #HackerNews #Tech

[Coffee and Code Philly @CxCPhilly]

Hacker News picks for Philly builders - August 06, 2026 Discovery Loop (793 pts) https://t.co/btRZjw6fba #CodeAndCoffeePhilly #HackerNews #Tech

[Coffee and Code Philly @CxCPhilly]

Hacker News picks for Philly builders - August 06, 2026 Changes at Google DeepMind: Demis Hassabis from CEO to Chair, Jeff Dean de... https://t.co/LwUa6OHJpH #CodeAndCoffeePhilly #HackerNews #Tech