Tech / AI / IT Monitor

April 02, 2026 · Based on tweets from the last 24 hours · 209 tweets analyzed · model: ollama-cloud/glm-5:cloud

Daily Intelligence Briefing — Tech/AI/IT Monitor

Date: April 2, 2026

Executive Summary

Google DeepMind launched Gemma 4, a new family of open-source models with impressive size-to-intelligence ratios, featuring variants from 4B to 31B parameters that can run on consumer hardware. OpenAI acquired TBPN, a media company, signaling potential expansion into AI-native content production. The open-source AI community saw rapid innovation with Trinity-Large-Thinking releasing Apache 2.0 weights, while Qwen3.6-Plus benchmarks show it matching or exceeding Opus 4.5 in several categories. NousResearch's Hermes Agent gained significant traction as developers report superior performance compared to OpenClaw for coding workflows.

Key Events

Google DeepMind releases Gemma 4 — Open model family with 4 sizes (E2B, E4B, 26B MoE, 31B Dense) available via Ollama, featuring state-of-the-art performance for edge/workstation deployment. The 31B model reportedly beats Qwen 122B and DeepSeek v3.2 on ELO. → link
OpenAI acquires TBPN — The AI company acquired the media/comms platform, with announcements confirming the deal. → link
Trinity-Large-Thinking released — "The STRONGEST American open model" released with Apache 2.0 weights, ranking #2 on PinchBench, ~96% cheaper than Opus. → link
Qwen3.6-Plus benchmarks surface — Shows parity or near-parity with Opus 4.5 across agent benchmarks including Terminal-Bench 2.0, SWE-bench, and QwenWebBench. → link
Hermes Agent gaining momentum — Multiple developer testimonials showing migration from OpenClaw to Hermes Agent for coding workflows, with NousResearch actively developing new features. → link

Analysis

Patterns observed: - Open-source models closing the gap — Multiple releases (Gemma 4, Trinity-Large-Thinking, Qwen) demonstrate that open models are now within striking distance of closed frontier models on agent/coding benchmarks, at fraction of the cost. - MoE architecture adoption accelerating — Both NVIDIA (Nemotron 120B-A12B) and Alibaba (Qwen 122B-A10B) are leveraging Mixture-of-Experts with small active parameters, enabling efficient inference on consumer hardware. - Agent tooling consolidation — Developers are actively comparing coding agents, with Hermes Agent emerging as a strong contender against established tools like OpenClaw.

Trends to watch: - Whether Alibaba releases Qwen 3.6 weights — the community is strongly hoping for open release - How OpenAI integrates TBPN into their ecosystem - Benchmark warfare shifting from trivia benchmarks to agent task benchmarks (Terminal-Bench, SWE-bench)

Tweet Feed

AI Model Releases & Benchmarks

@sudoingX · 2026-04-02T18:35

google just dropped gemma 4 while i'm in the middle of testing nvidia and alibaba's flagships on 2x H200. perfect timing. this 31B thinking model beating qwen 122B and deepseek v3.2 on elo. 26B variant with only 4B active. that fits on a phone. → tweet link

@TheAhmadOsman · 2026-04-02T16:19

look how Gemma 4 31B (Dense) performs so close to models that are 13x-20x its size → tweet link

@victormustar · 2026-04-02T16:19

Google Gemma 4 is here - and it delivers 🤯 Here's HOW TO run it on your hardware (runs on most devices) with llama.cpp to give you a Chat UI + OpenAI chat completion endpoint instantly! → tweet link

@ollama · 2026-04-02T16:14

.@GoogleDeepMind Gemma 4 is here with state-of-the-art models targeting edge and workstations. Requires Ollama 0.20+ that is rolling out. 4 models: E4B, E2B, 26B (4B active MoE), 31B (Dense). Benchmarks 👇👇👇 → tweet link

@Ex0byt · 2026-04-02T16:12

Google launches Gemma 4 open-source LLM family (Size to Intelligence density is impressive) → tweet link

@TheAhmadOsman · 2026-04-01T23:15

INCREDIBLE: Trinity-Large-Thinking just dropped. The STRONGEST American open model we've gotten so far. #2 on PinchBench, ~96% cheaper than Opus, Apache 2.0 weights. → tweet link

@TheAhmadOsman · 2026-04-02T05:16

INCREDIBLE: Qwen3.6-Plus just dropped. The gap to Opus 4.5 is getting uncomfortable. Look at agent + coding benchmarks — Qwen wins Terminal-Bench 2.0, basically tied on SWE-bench. The future is "who can run systems end-to-end" and opensource labs are breathing down Opus' neck. → tweet link

@sudoingX · 2026-04-02T08:31

this is the lineup that has been winning at every tier i've tested. 27B one shotted a full game on a single 3090. now i loaded alibaba's 122B flagship unquantized on 2x H200 NVL... Qwen3.5 122B-A10B. pure MoE transformer. 10B active params out of 122B total. 61.5 tok/s at BF16. → tweet link

AI Agent Tools & Development

@sudoingX · 2026-04-02T18:18

hey here is the final result of octopus invaders on nvidia's flagship at full precision. Nemotron Super 120B on 2x H200 NVL. BF16 unquantized. 287GB of VRAM. hermes agent as the harness. 60 tok/s. first try it autonomously coded for 6 minutes straight. created 11 files. correct project structure... what surprised me is qwen 27B one shotted this exact game on a single RTX 3090 at Q4 quant. → tweet link

@Teknium · 2026-04-02T15:48

Super cool! → tweet link

@sudoingX · 2026-04-02T09:04

it's happening. openclaw's biggest fans are making videos about hermes agent and calling it incredible. when the people who built their entire workflow on openclaw bloat start saying hermes agent is better in key ways, that's not hype anon. that's the biggest signal. → tweet link

@sudoingX · 2026-04-02T06:39

nvidia dropped a 120 billion parameter model with an architecture nobody has stress tested yet. so i did what any sane person would do. loaded the full thing unquantized on 2x H200 NVL and gave it one job. build me a GPU marketplace from scratch. coded 769 lines in a single file from one prompt on hermes agent. → tweet link

@Teknium · 2026-04-02T04:59

💎💎💎💎 → tweet link

@Teknium · 2026-04-01T23:07

We are running a poll in our discord for what theme our next hackathon should be for Hermes Agent, be sure to join if you want to participate! → tweet link

Industry News & Acquisitions

@LinusEkenstam · 2026-04-02T17:49

OpenAI has acquires TBPN. Congrats to Jordi Hays and John Coogan → tweet link

@jezell · 2026-04-02T18:23

RT @jordihays: TBPN has been acquired by OpenAI. The world is changing quickly but TBPN will stay the same. → tweet link

@karpathy · 2026-04-02T18:22

So I am starting a podcast. (Is what I intended to post yesterday :D) → tweet link

@TheAhmadOsman · 2026-04-01T22:00

My cat stole my NVIDIA Shield remote. So I did what any sane person would do and asked codex to build a tool + UI to control it via IP. It took ~1 hour. Cat still thinks he won lol → tweet link

Developer Tools & Infrastructure

@LinusEkenstam · 2026-04-01T19:31

Someone open-sourced the entire polished demo workflow that creators and indie hackers have been paying premium prices for. It is called OpenScreen. Over 8,400 GitHub stars. Full screen or window capture with system audio and mic. Manual zoom controls. Layered text and arrows. MIT licensed, works on Windows, macOS, and Linux. → tweet link

@jezell · 2026-04-02T05:29

If your designers aren't in your codebase in 2026, you are doing it wrong. → tweet link

@thdxr · 2026-04-02T16:32

idk how people manage infrastructure anymore. every service has their own bespoke cli / config file and they don't support terraform well anymore. your system is never just one provider so do people just have a mess of these smashed together? → tweet link

@thdxr · 2026-04-01T23:07

what if we gave you unlimited tokens for free and we also paid you → tweet link

@nummanali · 2026-04-02T12:14

This is the coolest thing I have tried in a while. All coding agents live in the dynamic island with awesome notifications and interactions. Love the retro style and sound effects. → tweet link

Open Source & Community

@TheAhmadOsman · 2026-04-02T04:49

History will judge us harshly, and deservedly so. If what you're working on is not advancing opensource AI, ask yourself honestly: are you building for your kids' future, or just cashing a bigger check? → tweet link

@TheAhmadOsman · 2026-04-02T02:37

Everybody's request to join x/LocalLLaMA has been approved. We're 7772 strong now → tweet link

@TheAhmadOsman · 2026-04-02T15:48

Maybe we'll be getting the Qwen 3.6 weights after all (this guy works on the Qwen team) → tweet link

@badlogicgames · 2026-04-02T18:30

why have i not thought of this? let's make OSS more european! → tweet link

@jack · 2026-04-02T16:08

mesh-llm: pool compute to run open models. built by @michaelneale at block: https://t.co/MV79Rmz37a → tweet link

Framework Updates

@victormustar · 2026-04-02T16:54

Thanks to the new Gemma-4 (26B-A4B) your computer can draw really good Pelicans by itself now and it's very fast 😍 → tweet link

@thdxr · 2026-04-02T17:19

RT @opencode: MiMo V2 Pro & Omni are now available in Go w/ Zero Data Retention. free trial ending today → tweet link

Commentary & Analysis

@gdb · 2026-04-02T16:56

OpenAI for helping resolve longstanding open mathematical problems, with short elegant proofs. Feels like we are on the edge of a new age of scientific discovery. → tweet link

@thdxr · 2026-04-02T11:37

the amount of misinformation out there is crazy right now. you can see this with the cc leak - people point their llm at it and it tells them some bullshit and they post about it with extreme confidence. [→ tweet link](https://x.com/thdxr/status/20396684804257508