Executive Summary
The past 24 hours reveal significant momentum in open-source AI with Hermes Agent reaching 50K GitHub stars and growing its community to 4,500+ members, positioning itself as a compelling alternative to OpenClaw. Google DeepMind's Gemma 4 models continue receiving praise as best-in-class for their size, running efficiently on consumer hardware like RTX 5090s. The AI infrastructure supply chain spotlighted Ajinomoto's ABF (Build-Up Film) as a critical monopoly with 95% global share in semiconductor packaging. Meanwhile, skepticism grows around Anthropic's "Mythos" release, with researchers questioning benchmark validity and suggesting it may be a looped language model rather than a genuine breakthrough.
Key Events
-
Hermes Agent crosses 50K GitHub stars with exponential growth trajectory, gaining 25K stars in just 8 days, as the community positions it as a token-efficient alternative to OpenClaw with native Linux support and WeChat integration. → link
-
Gemma 4 earns significant praise from the open-source community as "truly the best open model for its size by a long landslide," with 31B Turbo variant running at 51 tok/s on single RTX 5090 and 1244 tok/s batched. → link
-
Ajinomoto's ABF semiconductor film exposed as critical AI infrastructure with 95% global market share and 100% in high-end AI data centers; activist investor Palliser pushes for 30% price increases as ABF content in Nvidia's Vera Rubin is 4x that of H100. → link
-
Skepticism mounts around Claude Mythos with researchers claiming they replicated findings using GPT-5.4, suggesting Mythos may be a "looped language model" and criticizing benchmark gaming. → link
-
tinygrad merges Bonsai-8B support for 1-bit LLMs in just 5 lines of code, demonstrating the framework's abstraction quality. → link
-
VibeJam game development contest continues with $35,000 in prizes, showcasing AI-assisted game development progress with 21 days remaining. → link
-
Codex CLI 0.119.0 released with markdown copy and new features from community contributors. → link
Analysis
Patterns: The open-source AI momentum is accelerating, with Hermes Agent and Gemma 4 emerging as credible alternatives to proprietary solutions. A clear bifurcation is forming between "bloat" tools (criticized for token inefficiency) and streamlined local-first agents. The semiconductor supply chain analysis reveals hidden infrastructure dependencies—the Ajinomoto ABF monopoly mirrors other critical chokepoints in AI hardware.
Escalation/De-escalation: Tensions are rising between competing agent frameworks (Hermes vs OpenClaw), with aggressive migration messaging. Benchmark credibility is under fire as the community becomes more skeptical of marketing claims, particularly around Mythos. The "vibe" ecosystem (vibe coding, vibe hacking, vibe security) is emerging as a new paradigm in AI-assisted development.
What to Watch: - Hermes Agent's trajectory toward 100K stars by April 14 "Jam" event - Replication attempts of Mythos using existing models - Ajinomoto's response to activist pressure on ABF pricing - RTX 3090/5090 benchmark comparisons between Qwen 27B, Gemma 4, and Carnic-27B
Tweet Feed
AI Models & Releases
@Ex0byt · 2026-04-11T14:55
Gemma4 is a national treasure you're all still sleeping on. GoogleDeepMind threw OSS AI a lifeline with the Gemma 4 class models. A week in and I can confidently attest, truly the best open model for its size by a long landslide. Connect your agents, thank me later. → tweet link
@tinygrad · 2026-04-11T10:19
Just merged an external PR for Bonsai-8B support (1 bit LLM). Because tinygrad has the correct abstractions, it was 5 lines. https://t.co/BLljWDANgq https://t.co/GlXWqPbYg5 → tweet link
@Ex0byt · 2026-04-11T05:47
The (Mythos) Kraken is out. (MLX) https://t.co/zvuYKFVq9s (GGUF) https://t.co/wzNgjhGIK3 https://t.co/ejWbR3PTJz → tweet link
@sudoingX · 2026-04-10T21:03
the best model to run on a single rtx 3090 is qwen 27b dense. the rest i have tested don't even come close.
and this is BEFORE i benchmark Carnic-27b and gemma 4 dense. so the throne could shift. nothing has moved it yet. → tweet link
Hermes Agent & AI Agents
@sudoingX · 2026-04-11T07:18
three weeks ago i said 100K stars was the target. and today holy shit hermes agent just crossed 50K and the curve is still going vertical.
look at the chart. september to march was gradual. march to april went exponential. the last 25K stars landed in 8 days. → tweet link
@Teknium · 2026-04-11T10:07
Hermes was built linux first! It has some special capabilities if running on a mac (like accessing your notes) but, generally, it performs super well on a Linux OS! → tweet link
@sudoingX · 2026-04-11T07:35
wow from zero to 4,500+ members in less than a month.
the hermes agent community is the fastest growing AI agent room on x right now. no ads, no giveaways, no gatekeeping. just builders helping builders get their setups running. → tweet link
@Teknium · 2026-04-11T07:41
Tip of the day - use hermes skills browse or hermes skills search to search and browse across many different popular skills hubs, built right into Hermes Agent!
We have integrated skill hubs like ClawHub, https://t.co/O3mPXXRZr6, and more. → tweet link
@sudoingX · 2026-04-11T07:57
this is the migration pattern in one post. i see it every day. openclaw bloat burns tokens that are literally money burning for free.
you deserve better tools for your cognition. drop the bloat. stop getting frustrated by api bills. make the move before your monthly budget is torched in days.
hermes agent is the answer that's been under your nose the whole time. → tweet link
AI Infrastructure & Hardware
@TrungTPhan · 2026-04-11T17:08
ABF currently has a 95% global share in ABF and 100% share in high-end AI data center applications.
While mission-critical for high-density AI packaging, ABF still represents less than 0.1% of a GPU's sale price.
As GPU chips advance, ABF has become even more important. There is 4x more ABF in Nvidia's Vera Rubin vs. H100 chip. → tweet link
@TheAhmadOsman · 2026-04-10T20:40
Who wants to see a DXG Spark vs a Mac mini M4 64GB comparison?
Which model do you wanna see running & benchmarked? → tweet link
@sudoingX · 2026-04-10T19:21
a few days ago my laptop kept freezing and i posted about it.
turns out the universe had already been answering with hardware. nvidia shipped me a 128gb dgx spark to thailand (stuck in customs). a builder in the community gave me h200 access. someone sent 2.15 ETH ($4,500) for a new laptop 24 hours after the freezing post. ROG strix scar 18 with a mobile 5090 is on my desk today. → tweet link
Developer Tools & Coding
@steipete · 2026-04-11T00:47
RT @fcoury: Codex CLI 0.119.0 is out and I'm excited about two features I worked on:
📋 Ctrl+O now copies the last agent response as markdown... → tweet link
@TheAhmadOsman · 2026-04-11T13:31
LLM infra right now is like Linux in the 90s
We're still early and there is a lot of work to do → tweet link
@TheAhmadOsman · 2026-04-10T19:04
Bash (tool / philosophy) plays a major role in how useful AI Agents are today
Take a weekend and learn
unix basics (cd, mv, cp, nano, etc) git (branch, merge, rebase) ssh, ports & firewalls 101
Highest leverage multiplier for agents → tweet link
@thdxr · 2026-04-11T17:21
i'm pretty proud of the "37 prompts just to turn a button red" line → tweet link
@levelsio · 2026-04-11T10:44
✉️ Vibe coded a POP3 and SMTP text-based email app for https://t.co/1uskGoglfJ that uses the dial-up modem to check and send emails and it worked on the first try → tweet link
Benchmarking & Mythos Skepticism
@jezell · 2026-04-11T04:42
RT @kannthu1: UPDATE: We were able to replicate the Mythos findings using existing models (GPT5.4)
Writeup coming early next week, no BS p... → tweet link
@jezell · 2026-04-11T04:42
RT @RokoMijic: Mythos was a marketing exercise → tweet link
@jezell · 2026-04-11T05:51
Vibe Hacking is a think now. So it's gonna be the Vibe Hackers vs the Vibe Security suites now eh? → tweet link
@jezell · 2026-04-11T06:34
RT @wholyv: CLAUDE MYTHOS MIGHT JUST BE A TOTAL HYPE.
a very great hacker just said Mythos is not as good as they all are hyping it up as... → tweet link
AI Video & Creative Tools
@LinusEkenstam · 2026-04-10T20:23
A year ago AI video was 5-second clips. today it's a full episode. Zypher a new Higgs original. Adult k-pop meets mechs, a glimpse into the future of entertainment right here.
Seedance 2.0 → tweet link
@victormustar · 2026-04-10T19:14
RT @linoy_tsaban: It's out, and it comes with a demo🔥
Introducing Flux.2 [klein] Move ✨
specialized move & resize LoRA for Flux.2 [klei... → tweet link
Open Source & Community
@Ex0byt · 2026-04-11T04:36
Moving from open source