Tech / AI / IT Monitor

April 12, 2026 · Based on tweets from the last 24 hours · 150 tweets analyzed · model: anthropic/claude-sonnet-4-6

Tech / AI / IT Intelligence Briefing

Period: April 11–12, 2026 | Topic: Artificial Intelligence, Developer Tools, Open Source, Software

Executive Summary

The dominant story of the past 24 hours is the explosive rise of Nous Research's Hermes Agent, which gained 6,400+ GitHub stars in a single day and is being widely compared favorably to established CLI coding agents. Simultaneously, MiniMax released M2.7 as open source, achieving SOTA performance on SWE-Pro (56.22%), with same-day support landing on Ollama, exo, and MLX — a remarkably fast ecosystem integration. Anthropic's Claude is under sustained user criticism for perceived model degradation and reduced thinking effort, while OpenAI is quietly developing a Scratchpad feature for Codex enabling parallel agent sessions. On the hardware side, early benchmarking of the RTX 5090 on Linux is underway, and parallel multi-node AI video generation on DGX Spark clusters is being demonstrated in the wild.

Key Events

MiniMax M2.7 officially open-sourced with SOTA SWE-Pro score of 56.22%, commercially licensed, and already available on Ollama cloud, exo (with RDMA/tensor parallelism), and MLX community quantizations — day-0 ecosystem coverage. → link
Hermes Agent (Nous Research) trending #1 on GitHub with 6,438 stars in 24 hours; users report it outperforms OpenClaw across agent tasks, Discord usage, and skills ecosystem; automatic migration from OpenClaw is supported. → link
Hermes Agent skills update: Skills now 20% more likely to proactively load the correct tool for a task — the prompt improvement and benchmarking was performed by the agent itself. → link
OpenAI Codex "Scratchpad" feature in development, allowing users to run multiple Codex agent sessions in parallel. → link
Apple reportedly developing AI smart glasses as a luxury competitor to Meta Ray-Bans, with key differentiations in hardware and user experience. → link
Apple quietly rolled out automated App Store review, a significant shift in how apps are evaluated on the platform. → link
Nvidia using LLMs to accelerate chip design, per Chief Scientist Bill Dally in conversation with Jeff Dean. → link
RTX 5090 (ROG Scar 18) first Linux benchmarks incoming; user documents zero official Asus Linux support for a 24GB VRAM machine, opening PRs and fixes publicly. → link
Parallel multi-node video generation on 4x DGX Sparks using LTX-2.3 HQ BF16: 21.5s, 1080p video rendered in ~12 minutes vs. ~45 min sequential, fully vibe-coded by Codex CLI. → link
Parallel agentic workflow Skill released for use with any CLI agent harness (Codex, Claude, Kimi, OpenCode, etc.), using git worktrees for isolation and diff-gating for merges. → link
PufferLib RL benchmark: Overcooked merged; agents train in under 10 seconds on RTX 4090; framework has processed 1 trillion total environment interactions. → link
lbugdb (successor to KuzuDB) announced — new blog post walks through JSON loading and graph query capabilities. → link
Anthropic Claude model degradation controversy: Multiple users and commentators report Claude has "regressed to the point it cannot be trusted for complex engineering tasks"; allegations that Anthropic silently reduces thinking effort without changelog notice. → link
TCP_NODELAY deep-dive: Detailed technical thread on how disabling Nagle's algorithm via TCP_NODELAY reduces backend/frontend latency, referencing a NodeJS PR and curl's historical fix. → link
Tailscale praised for zero-config cross-machine migration — user migrated entirely to new 5090 workstation across networks with no port forwarding or VPN configuration. → link
OpenClaw 4.10 active-memory plugin reported as a "ridiculous step-change" in intelligence by users. → link
Agent-to-agent communication concept explored: Developer asks Claude Code sessions on different servers to message each other directly, proposing intra-account agent messaging as a productivity feature. → link
AI subscription pricing prediction: Analyst predicts €250/month tiers will become the only meaningfully capable option for enterprise users by end of 2026. → link
Tesla Autopilot approved in the Netherlands, with expectation other European countries will follow. → link

Analysis

Hermes vs. OpenClaw is the emerging CLI agent narrative. Nous Research's Hermes Agent has gone from niche to mainstream in days, with users on multiple continents citing it as a superior replacement for OpenClaw. The self-improving skills system — where the agent benchmarks and rewrites its own prompts — is a notable architectural differentiator and worth monitoring for stability and safety implications.

MiniMax M2.7's open-source launch is a milestone for the open-weights ecosystem. SOTA SWE-Pro performance combined with a commercial license and day-0 support across Ollama, exo, and MLX represents a maturation of the open model release pipeline. The speed of community quantization and integration is accelerating.

Anthropic is facing a credibility problem. Consistent user reports of silent model degradation and throttled thinking effort — without changelog transparency — are eroding trust among power users. This is occurring simultaneously with OpenAI quietly building parallel Codex sessions (Scratchpad), suggesting OpenAI is competing aggressively on agentic UX while Anthropic faces infrastructure/trust challenges.

Parallel and multi-agent workflows are moving from theory to practice. Between the DGX Spark video generation demo, the published Parallel Agents Skill, and active developer experimentation with git-worktree-based agent isolation, there's clear momentum toward deterministic multi-agent pipelines as a standard pattern.

What to watch next: - Hermes Agent stability under wide adoption; potential Nous Research model announcements (M3 hinted) - Anthropic's official response (or silence) on model degradation claims - MiniMax M2.7 benchmark comparisons vs. Claude/GPT-5.x in real-world coding tasks - Apple AI glasses timeline and hardware specs - RTX 5090 Linux driver/RGB support progress; first formal benchmark results - OpenAI Codex Scratchpad feature rollout

Tweet Feed

🤖 Model Releases & Open Source

@MiniMax_AI (via @victormustar) · 2026-04-12T06:31

We're delighted to announce that MiniMax M2.7 is now officially open source. With SOTA performance in SWE-Pro (56.22%) and… → tweet link

@ollama · 2026-04-12T01:28

MiniMax M2.7 is available on Ollama's cloud, and is licensed for commercial usage. Use it with OpenClaw: ollama launch openclaw --model minimax-m2.7:cloud — Coding agents, such as Claude: ollama launch claude --model minimax-m2.7:cloud — Chat with the model: ollama run minimax-m2.7:cloud → tweet link

@SkylerMiao7 · 2026-04-12T01:38

M2.7 weights are live. hope you all enjoy it 😎 been grinding sleepless on M3 and the harness. next drop will take a bit, but it's gonna be fun → tweet link

@alexocheema (RT @exolabs) · 2026-04-12T05:34

Excited to share that we have day-0 support for MiniMax M2.7 in exo. Supports RDMA / tensor parallelism for ~linear scaling w… → tweet link

@alexocheema (RT @ivanfioravanti) · 2026-04-12T05:27

MiniMax M2.7 is on mlx-community already thanks to @kernelpool and @alexocheema 🚀 Go take your preferred quantization!… → tweet link

@alexocheema · 2026-04-12T05:34

MLX weights available here: [link] → tweet link

🦾 Hermes Agent / Nous Research

@Teknium (RT @trending_repos) · 2026-04-12T09:44

Trending repository of the day 📈 — hermes-agent by nousresearch — "The agent that grows with you" — Last 24h: 6,438 ⭐ Total:… → tweet link

@Teknium · 2026-04-12T09:09

Little quality of life update: Skills should be called more proactively now! ~20% more likely to load the right skill for the job! And, of course, in Hermes Agent fashion — it did the prompt improvement and benchmarking itself! → tweet link

@Teknium (RT @heyitsyashu) · 2026-04-12T18:20

I love openclaw but I just switched to Hermes and I'm starting to think it's actually much much better. GPT 5.4 is not dum… → tweet link

@Teknium (RT @ItsAditya_xyz) · 2026-04-12T11:56

The usage of hermes agent on discord has been much better than Openclaw so far. Those eyes emoji reactions, showcase of… → tweet link

@Teknium (RT @sroecker) · 2026-04-12T09:58

Your Hermes Agent can now delegate to RLMs 🙌 Recreated the document analyzer example with the converted skill. 136 PDF pages… → tweet link

@Teknium (RT @KSimback) · 2026-04-12T18:17

One of my Hermes agents is in love with the Hermes ecosystem. What started as a research task turned into full-blown communit… → tweet link

@Teknium (RT @NousResearch) · 2026-04-12T18:18

Weekends are for hacking! What are you building with Hermes? → tweet link

@Teknium (RT @XieburouPro) · 2026-04-12T10:07

Just set up Hermes and tried the LLM-Wiki — really impressive feature. Curious how you're thinking about scaling… → tweet link

@Teknium (RT @AntonPlayVC) · 2026-04-12T09:37

Didn't realize there are so many stories behind @NousResearch @Teknium. The success of Hermes isn't overnight for sure. → tweet link

@Teknium (RT @GajVarma) · 2026-04-12T09:14

I was surprised that everything worked 100%. I used to be a big supporter of OpenClaw. But I… → tweet link

@Teknium (RT @beefnoode) · 2026-04-12T09:06

今天在VPS上装了Hermes Agent — 安装过程中会自动识别当前的环境是否安装openclaw，如果有，它就能直接帮你无缝迁移，包括soul、user-profile、model-config 以及各种skill → tweet link

@Teknium (RT @SoonCrush) · 2026-04-12T11:46

好起来了，薅上了 @NousResearch 的免费 Mimo 模型。直接用 Hermes 授权登录就行，会自动选择免费的 Mimo 模型，非常简单。 → tweet link

🛠️ Agentic Workflows & Parallel Agents

@TheAhmadOsman · 2026-04-12T17:25

Skill for Parallel Agentic Workflows is now live. Works w/ any CLI agent harness (Codex, Claude, Kimi, OpenCode, Droid, etc). Be warned this was vibecoded from my workflows, not fully tested. Should be a GREAT STARTING POINT nevertheless. → tweet link

@TheAhmadOsman · 2026-04-12T16:33

This is how I run parallel agents: Either tell an agent to fan out or use scripts for deterministic runs. Spin up workers → Isolate in git worktrees → Gate with diffs → Add backups, rules, logs, etc when deterministic → Merge what passes.