Tech / AI / IT Monitor

April 07, 2026 · Based on tweets from the last 24 hours · 196 tweets analyzed · model: ollama-cloud/glm-5:cloud

Daily Intelligence Briefing: Tech / AI / IT Monitor

Date: April 7, 2026 (Last 24 Hours)

Executive Summary

GLM-5.1 emerged as the dominant open-weight AI model, achieving #1 ranking in open source on SWE-Bench and surpassing GPT-5.4, Opus 4.6, and Gemini 3.1 Pro with 58.4% score—available immediately on Hugging Face and Ollama. Hermes Agent from Nous Research gained significant traction with 300+ contributors, introducing self-improving memory systems and Karpathy's LLM-Wiki integration for research-heavy workflows. Anthropic's enterprise coding strategy appears vindicated with reported $30B ARR, while the local AI community focused heavily on optimizing consumer hardware (RTX 3090) for running powerful models locally. Gemma 4 (31B dense variant) discussions intensified with users reporting "step-change" improvements.

Key Events

GLM-5.1 released as top open-source coding model — Beat Opus 4.6, GPT-5.4, and Gemini 3.1 Pro on SWE-Bench Pro with 58.4% score; available as open-weight MIT license. → link
Hermes Agent gains momentum — Over 300 contributors, self-improving memory architecture, and LLM-Wiki integration for knowledge base creation shipped. → link
Anthropic reaches $30B ARR — Revenue grew from $1B to projected $30B in 15 months, driven by enterprise coding adoption. → link
Gemma 4 dense 31B model praised — Users describe it as a "step-change" in performance, with ongoing work on PRISM-PRO-DQ quantized variants. → link
Ollama launches GLM-5.1 cloud integration — Immediate availability through subscription model for OpenClaw and Claude Code workflows. → link
Chrome adds vertical tabs — Major browser ux shift for power users managing many tabs. → link
NVIDIA PersonaPlex 7B released — Open-source real-time conversational voice model with 18x faster interruptions than Gemini Live. → link

Analysis

Patterns: - Open-weight models closing the gap: GLM-5.1's benchmark dominance signals that open-source models are no longer trailing — they're leading in specific categories. The MIT licensing approach is becoming competitive leverage. - Agent infrastructure consolidation: Hermes Agent's rapid community adoption (300+ contributors) suggests the open-source community is coalescing around fewer agent frameworks rather than fragmenting. - Consumer GPU relevance persists: RTX 3090 (7-year-old hardware) remains viable for running modern models, extending hardware investment value for developers. - Enterprise coding AI dominance: Anthropic's revenue trajectory confirms B2B/enterprise focus beats consumer chat interfaces for AI monetization.

Escalation signals: - Multiple major model releases in single week (Gemma 4, GLM-5.1, MiniMax-2.7 teased) - Agent memory/skill systems becoming pluggable infrastructure - Tinygrad racing multiple frontier models for support

Watch next: - MiniMax-2.7 open weights release (confirmed "coming soon") - Gemma 4 PRISM-PRO-DQ quantized variant finalization - Hermes Agent community tool integration bridges

Tweet Feed

AI Model Releases & Benchmarks

@louszbd · 2026-04-07T17:43

we open-sourced glm-5.1 agents could do about 20 steps by the end of last year. glm-5.1 can do 1,700 rn. autonomous work time may be the most important curve after scaling laws. glm-5.1 will be the first point on that curve that the open-source community can verify with their own hands. hope y'all like it^^ → tweet link

@victormustar · 2026-04-07T16:24

GLM-5.1 weights are on Hugging Face - probably the best open weights model in the world right now :) https://t.co/hjt26kG0m8 → tweet link

@victormustar · 2026-04-07T18:21

GLM-5.1 is now available in HuggingChat https://t.co/UK3Q9eo1FV Getting excellent vibes from it, I recommend this prompt: "generate a beautiful modern landing page about a wild monkey sanctuary (generate 10 beautiful images in parallel that you'll use)" → tweet link

@Ex0byt · 2026-04-07T18:15

G4-It (31B Dense) Is definitely a step-change. Honestly, I am shook. → tweet link

@Ex0byt · 2026-04-07T17:24

Dilemma: Gemma-4-PRISM-PRO-DQ (Dense 31B) is otherwise ready to ship — but it feels too powerful to responsibly drop as-is. Quantize it down, or something else? Genuinely open to suggestions. → tweet link

@Ex0byt · 2026-04-07T16:49

What an exciting week of Open Source! First Gemma4, now GLM-5.1 Is out. Review and GLM-5.1-PRISM-PRO-DQ Incoming. → tweet link

@victormustar · 2026-04-07T08:38

MiniMax-2.7 open weights confirmed and coming super soon - this is going to be a very big one 🔥 https://t.co/qrRuBN0p4N → tweet link

Hermes Agent & OpenClaw Development

@Teknium · 2026-04-07T06:56

Shout out to the over 300 people who've contributed to Hermes Agent to make it what it is! https://t.co/fMxNI7vYym → tweet link

@Teknium · 2026-04-07T18:21

MIMO V2 PRO FREE FOR TWO WEEKS - SIGN UP FOR NOUS PORTAL AND UPDATE HERMES AGENT TO ACCESS!! → tweet link

@sudoingX · 2026-04-07T13:45

let me be very clear. i did not build hermes agent. hermes agent is built by @Teknium and @NousResearch with 300+ contributors who deserve recognition. what i am is a contributor, a tester, and the loudest fan in the room with some PRs merged and every benchmark i run goes through hermes agent. the community is something i helped grow but the codebase is teknium's work. credit where credit belongs. always. → tweet link

@sudoingX · 2026-04-07T17:03

if openclaw is still your daily coding agent in april 2026 you're not going to make it. → tweet link

@sudoingX · 2026-04-07T10:01

it's happening. people are telling me to drop it now. i get it. but i've had 3 PRs merged into hermes agent and i know what teknium's codebase looks like. it's clean. it's tested. his style is what i learned from those merges. so i'm resisting every temptation to rush this and doing the boring work first. because when this lands on the maintainer's desk it should be ready to merge. the feature works. now i make it worthy of the repo. → tweet link

@Teknium · 2026-04-07T04:28

Oh shit I'm in a video game now thanks to Hermes Agent 🤯🤯🤯 → tweet link

@sudoingX · 2026-04-07T16:16

its been about 48 hours now and the tool calls are firing. i'm watching an LLM call tools on my machine through a bridge nobody has built before. this will be on hermes agent for everyone once everything is clean and codebase worthy. → tweet link

Local AI & Hardware Optimization

@sudoingX · 2026-04-07T16:50

anyone with an RTX 3090 should be paying attention. this machine is not just a gaming setup in 2026. it can host an intelligence that one-shotted an entire game on a single card that 2x H200 at full precision could not. get your hands dirty. compile llama.cpp with CUDA. run qwen 3.5 27B dense Q4_K_M. undisputed on 24GB. the best model for that card and it's not even close. i haven't tested gemma 4 yet but it's on the queue. the 3090 era is far from over. → tweet link

@TheAhmadOsman · 2026-04-07T05:00

the basics: PCIe lanes, or the highways GPUs use for data transfer [...detailed PCIe guide covering Gen 3-6, bandwidth calculations, multi-GPU inference/training considerations...] → tweet link

@TheAhmadOsman · 2026-04-07T22:25

Just some numbers so you don't get misled RTX 3090 (7 years old) 24GB VRAM Bandwidth: 936.2 GB/s Bi-directional NVLink 112GB/s RTX PRO 4000 24GB VRAM Bandwidth: 672 GB/s No Bi-directional NVLink need 32 Gen. 5 PCIe Lanes to pool 2 at 64GB/s → tweet link

@TheAhmadOsman · 2026-04-07T01:53

Most people think "more memory = better AI box" and that's why their setups feel slow Local AI hardware isn't about one spec it's about tradeoffs Here's how capacity, bandwidth, and reality actually work ↓ → tweet link

@TheAhmadOsman · 2026-04-07T22:04

MoE models such as Gemma 4 26B-A4B and Qwen 3.5 35B-A3B prefer Memory Size > Memory Bandwidth Dense models such as Qwen 3.5 27B and Gemma 4 31B prefer Memory Bandwidth > Memory Size → tweet link

Developer Tools & Infrastructure

@ollama · 2026-04-07T16:39

GLM-5.1 is here! Try it on OpenClaw🦞🦞🦞 ollama launch openclaw --model glm-5.1:cloud Claude Code ollama launch claude --model glm-5.1:cloud Chat with the model ollama run glm-5.1:cloud → tweet link

@thdxr · 2026-04-07T01:42

opencode server now can run under nodejs as we're no longer using any bun specific apis we're working on a larger refactor now of its internals as we work towards a 2.0 will have nice updates to the plugin and sdk apis → tweet link

@badlogicgames · 2026-04-07T17:47

RT @calebfahlgren: Starting today, Agent Traces are native on @huggingface with support for Claude Code, Codex, and Pi 🤖 💽 Auto detection… → tweet link

@TrungTPhan · 2026-04-07T18:01

Chrome browser now has vertical tabs. Absolute game changer to organize the 146 articles and 78 PDFs I'll never read, along with the 69 Google Doc essays I'll never finish writing. → tweet link

@levelsio · 2026-04-07T14:21

💾 Made a DOS-style text-based user interface (TUI) today First app I made is Notepad (or should it be called EDIT.EXE?) and you can type in it and it has a localStorage-based file system with folders and everything! → tweet link

Anthropic & AI Industry News

@LinusEkenstam · 2026-04-07T12:46

Anthropic casually passed $30bn in ARR or in plain english, $20bn in net new annual business since end of 2025 Now they secured a whole batch of next level TPU's from Google. I guess there is no such thing as bad PR. → tweet link

@FinansowyUmysl · 2026-04-07T17:18

Anthropic przeszedł drogę od 1 mld $ przychodów do 30 mld $ w zaledwie 15 miesięcy. AI miało się nie nadawać do programowania, a okazuje się, że każda większa firma chce korzystać z modelów AI do programowania. I tutaj zdominował rynek Anthropic. → tweet link

@FinansowyUmysl