Executive Summary
The past 24 hours in AI and tech have been dominated by heated competition between AI platforms and a growing emphasis on local/decentralized AI solutions. OpenAI has made aggressive moves to attract enterprise coding customers with free Codex trials while simultaneously facing renewed confidence from developers following Anthropic's restrictions on programmatic Claude access. Nvidia's Nemotron 3 Nano Omni model is gaining traction in local AI circles as an underrated open-source option, while new research on training speed optimization (Superposition) and GUI agent memory systems (EAM) signal continued innovation in agent frameworks. The developer community is increasingly vocal about the necessity of agent-accessible APIs and protocols, with debates around OpenClaw versus Hermes Agent highlighting the evolving landscape of local AI tooling.
Key Events
- OpenAI Offers Free Codex Trials — Sam Altman announces 30-day free trial (2 months) for companies switching to Codex, positioning it as the "best AI coding product" amid intensifying competition. → link
- Anthropic Reverses Claude API Restrictions — Following developer backlash to programmatic subscription restrictions, Anthropic announces monthly credits equal to plan price for third-party apps using Agents SDK, enabling OpenClaw, Hermes, and T3 Code to utilize Claude allowances. → link
- Nvidia Nemotron 3 Nano Omni 30B-A3B Gains Local AI Traction — Developers praise Nvidia's open-source multimodal model as highly underrated, featuring reasoning, video understanding, and image vision in a MoE architecture that fits on consumer hardware like DGX Spark. → link
- Superposition Training Paper Released — Researchers introduce potential method for multiplying pre-training speed, representing significant efficiency advancement in model training. → link
- Executable Agentic Memory (EAM) Breakthrough — New framework gives GUI agents persistent knowledge graphs, achieving +19.6% accuracy improvement over SOTA baselines with 6x fewer tokens. → link
- npm Supply Chain Attack Response — Developer @kunchenguid shares hardening strategies against ongoing npm supply chain attacks, advising enforcement of third-party version publication delays. → link
- TinyBox Pro v2 Launched — tinygrad releases new hardware lineup starting at $45k with RTX 5090s and Blackwell GPU options up to 768GB RAM. → link
Analysis
Platform Competition Intensification: The Anthropic reversal represents a significant de-escalation after what many developers perceived as a platform lock-in strategy. This shift validates the developer community's influence and suggests sustainable AI platforms must maintain openness. OpenAI's aggressive Codex positioning with free trials indicates a price war in the enterprise coding assistant space.
Local AI Momentum: Multiple threads championing Nvidia's Nemotron and Hermes Agent suggest a growing segment of builders prioritizing local inference, privacy, and cost control over frontier model performance. The discourse around "vibe-coded dashboards" versus robust solutions indicates maturation of developer expectations.
Agent Infrastructure Maturation: The emphasis on MCP protocols, CLI/API accessibility, and computer use capabilities reflects industry recognition that AI products must now serve both human users and agentic systems. This represents a fundamental shift in product design philosophy.
What to Watch: Google's I/O announcements (Monday), continued Hermes Agent vs OpenClaw competition, EAM framework adoption, and whether Anthropic's credit system satisfies developer requirements.
Tweet Feed
AI Model Releases & Research
@Teknium · 2026-05-13T18:42
Check out our researchers' latest paper that introduces Superposition, a potential path to multiplying training speed during pre-training. → tweet link
@Ex0byt · 2026-05-13T14:21
Will need to try this with JARVIS... new paper drop: Executable Agentic Memory (EAM) gives GUI agents a map. Instead of re-reading every screen from scratch, the agent builds a knowledge graph of what it's seen and what works, then navigates it. +19.6% accuracy on AndroidWorld vs UI-TARS-7B, matches closed cloud at 6x fewer tokens, -2.8s average per step → tweet link
@victormustar · 2026-05-13T09:35
Quite excited about llama-eval, a proposed eval tool for llama.cpp. Could be a nice step toward more comparable community evals 🎉 → tweet link
@sudoingX · 2026-05-13T07:13
nobody is talking about how good nemotron 3 nano omni 30b-a3b actually is on local. very underrated. multimodal, reasoning, video understanding, image vision, all shipped in one open source release by nvidia. moe architecture 30b total params, 3b active per token, q8 is near lossless and fits comfortably on a single dgx spark with room to breathe. → tweet link
@Ex0byt · 2026-05-12T19:32
GLM-5.1-PRISM-Pro acting all shady. → tweet link
@sudoingX · 2026-05-13T09:43
RT @sudoingX: hey anon, let me tell you this again because you only listen when repeated. qwen 3.6 27b dense q4 is the top king on a single… → tweet link
AI Coding & Developer Tools
@sama · 2026-05-13T18:17
i get some anxiety not using the smartest-available model/settings. but sometimes i dont mind if it's really slow. i wonder if we should focus more on a price/speed tradeoff relative to a price/intelligence tradeoff. → tweet link
@sama · 2026-05-13T18:13
codex is the best AI coding product and we want to make it easy to try. for the next 30 days, we are giving companies that want to try switching over two months of free codex usage. → tweet link
@steipete · 2026-05-13T05:27
Codex was debugging a Telegram issue and needed a new token, so it used Peekaboo to open the Telegram Mac app, talked to botfather and just did it. Computer Use is amazing. → tweet link
@steipete · 2026-05-13T05:36
Kudos to Microsoft, they're helping to get OpenClaw ready for enterprises. → tweet link
@badlogicgames · 2026-05-13T16:30
RT @dsp_: We are thinking about deprecating Sampling, Logging and Roots in MCP. Let me know if you rely on these. → tweet link
@SkylerMiao7 · 2026-05-13T14:11
One subscription, everything unlocked. API, CLI, Agent. All models, shared credits. Agent Teams brings multi-agent parallel execution with adversarial quality gates. Mavis as your personal AI chief of staff. Built for long-running tasks. Open source incoming with M3 😎 → tweet link
@nummanali · 2026-05-13T13:38
GPT 5.5 has been post trained well. "I don't love adding an ignore note if we can avoid it" → tweet link
AI Platform Competition
@kunchenguid · 2026-05-13T18:11
it's official. Anthropic pulled the plug on ALL programmatic use of claude subscription. i've found myself increasingly bullish about OpenAI. a few key reasons - 1. Anthropic's only lead was on coding, and gpt 5.5 has flipped that already 2. OpenAI has voice, image, video 3. it took an Anthropic to see the "Open"-ness of OpenAI. Codex on the other hand remained an open platform for developers to build on → tweet link
@nummanali · 2026-05-13T17:12
Claude plans now are giving monthly credits equal to the plan price to use in 3rd party apps utilising the Agents SDK. That means OpenClaw, Hermes, T3 Code etc will all be able to draw from this allowance! Biggest U Turn that probably no one saw coming from Anthropic → tweet link
Local AI & Agent Frameworks
@sudoingX · 2026-05-13T11:17
if my agent can't use your product, you just lost a customer. before i even click your link, my agent has already gone through it. if there's no cli, no api, no way for an agent to interact with what you built i'm gone. the age of acceleration demands a new layer. your product doesn't just need a UI for humans anymore. it needs a protocol for agents. → tweet link
@sudoingX · 2026-05-13T06:38
most of you don't know how hard hermes agent is optimized for local AI at the system level. watch the full setup flow on screen. you paste an openai-compatible v1 endpoint, hermes auto-detects every model running behind it. hermes ships with per model parsers, prompt template auto-handling, tool call format detection per model architecture. cloud first harnesses bolt local support on top. hermes agent is local first from the architecture out. → tweet link
@sudoingX · 2026-05-13T06:38
if you're getting started on local AI, this is the harness you start with. try for yourself and find out. anyone serious about local AI lands here eventually. → tweet link
@sudoingX · 2026-05-13T06:38
buy a gpu. 3090, 4090, dgx spark, whatever fits your budget. tier doesn't matter. running your first local model does. the moment your first prompt lands with no api between you and the model, your brain rewires. → tweet link
@Teknium · 2026-05-12T22:52
RT @tonysimons_: This is a pretty massive update! Hermes can now control your computer. With any freakin' model your little heart desires… → tweet link
@Teknium · 2026-05-12T20:39
Being shown off in Hermes Jam Session right now 👀👀👀 → tweet link
@badlogicgames · 2026-05-13T16:26
RT @aidandcunniffe: Nobody really knows what works right now re: Coding Agent workflows. Nobody knows what a "software factory" looks like. → tweet link
@sudoingX · 2026-05-13T09:31
RT @sudoingX: let me say this out loud here: there is absolutely zero reason to use openclaw in may 2026. a general agent exists. hermes ag… → tweet link
Hardware & Infrastructure
@tinygrad · 2026-05-12T20:33
All products are now available to order! $45k for tinybox green with 5090s, and we have two variants of the tinybox pro v2, one with 5090s and one with RTX6000 Blackwells (that's a whopping 768 GB of RAM) → tweet link
@alexocheema · 2026-05-13T16:24
4 x M5 Max MacBooks clustered with RDMA: 512GB @ 2456GB/s, $20k, 560W, quiet. Find me a better deal, that I can buy today. → tweet link
@sudoingX · 2026-05-13T08:51
another hardware company just reached out today. if everything goes well, more hardware arriving soon. the lab keeps growing. → tweet link
@TrungTPhan · 2026-05-13T15:45
RT @bearlyai: Someone made a small hardware device for the desk that monitors Claude token usage. Calls it Clawdmeter. Kinda genius. → tweet link
@tinygrad · 2026-05-12T21:55
In before I can build it myself and save money. If you enjoy building as a hobby project and value your time at 0 or negative, you can. But if you just want a machine learning that works, I strongly doubt you can beat our prices. → tweet link
Software & Security
@kunchenguid · 2026-05-13T16:33
i just hardened most of the npm packages I own against the ongoing supply chain attack. using this thread to share what i did in case it's helpful for others. the core idea is simple: don't use any third-party package versions that are just published. enforce this with tools. → tweet link
@badlogicgames · 2026-05-13T16:46
People of https://t.co/oUoqqL9hAp. Do not https://t.co/dQsm09RJ6y via any method other than what's shown on the website and in the docs. E.g. we do not publish to brew and never will. Someone else did. We have zero control over what goes into the brew release. → tweet link
@jezell · 2026-05-13T16:58
RT @chdb_io: chDB v4.1.7 is out. Parquet files from S3 now read correctly, default memory limits are off so big queries no longer get capp… → tweet link
@badlogicgames · 2026-05-13T11:27
RT @heyandras: We made a fake repo with fake bounties, and the bots are applying fake PRs, so we know who is fake, and we can ban them from… → tweet link
@badlogicgames · 2026-05-12T21:56
RT @jxnlco: github was the man in the middle the whole time → tweet link
Industry Commentary & Analysis
@LinusEkenstam · 2026-05-13T13:14
This will catapult human computer interaction forward as it is a massive leap. Make sure you check out all the demonstrations. They really make you feel like Jarvis is incoming. → tweet link
@LinusEkenstam · 2026-05-13T12:28
I'm flying from Barcelona to Google I/O on Monday morning. Who else is going? Who wants to meet up there? Freaking hyped for monday ✨ → tweet link
@LinusEkenstam · 2026-05-13T14:24
If you want to understand what it takes to make great AI video, this breakdown from @PJaccetturo on the work of Zach @Gossip_Goblin is bringing fire → tweet link
@LinusEkenstam · 2026-05-13T09:05
This week is the week of: "This should have been made by Adobe, but wasn't". This is the second such product to be released in 48h. Ponder looks absolutely fantastic. → tweet link
@LinusEkenstam · 2026-05-12T20:28
This is the Midjourney contender we've been waiting for. K2. → tweet link
@ASalvadorini · 2026-05-13T18:03
Google A/I 2026 👇👇👇 #AI #GoogleIO → tweet link
@badlogicgames · 2026-05-13T14:17
RT @bentlegen: Relatedly, I don't see how any OSS maintainer can safely merge code they didn't read/vet → tweet link
@badlogicgames · 2026-05-13T11:26
recommended viewing. great points on LLMs and education. we don't need to solve content, we need to solve the motivation problem. → tweet link
@badlogicgames · 2026-05-13T16:53
Great primer from Sebastian. "Working Code Doesnt Lie" → tweet link
@levelsio · 2026-05-13T12:34
Tax and accountancy software and bookkeeping industry have perverse incentives to keep the tax code and bookkeeping hard so you have to keep paying a lot for their software and to hire them. If they let AI just do it via their APIs, their value would go to $0 fast! → tweet link
@levelsio · 2026-05-12T22:43
Oh ffs it just hallucinated everything 😂 It never actually reconciled any expense. And Xero says they'll NEVER add that via API because "ReGuLatOrY constraints". So I guess I have to go the browser way huh → tweet link
@iamdevloper · 2026-05-13T14:51
my theory on boomer ceos being opposed to remote work is they simply don't like their families → tweet link
@iamdevloper · 2026-05-13T12:57
- Avoid Kubernetes → tweet link
@FinansowyUmysl · 2026-05-13T05:06 (Polish language — tech sector analysis)
Mam wolną gotówkę i nie wiem gdzie ją wrzucić... Problem: AI tnie cenę software. Drastycznie. To, co kiedyś kosztowało subskrypcję 50$/mies., dziś masz w czacie za 20$ albo za darmo. Wygrywają ci, co sprzedają "łopaty": chipy, hardware, energia. → tweet link
@RealGeneKim · 2026-05-12T23:31
RT @bstaples: 1/ Yesterday I published a letter to our customers and investors about GitLab Act 2. The agentic era is the largest opportunity for GitLab and for the industry. → tweet link
AI Video & Content Creation
@LinusEkenstam · 2026-05-13T14:24
If you want to understand what it takes to make great AI video, this breakdown from @PJaccetturo on the work of Zach @Gossip_Goblin is bringing fire → tweet link
@LinusEkenstam · 2026-05-13T05:42
This video deserves more views. → tweet link
@juliarturc · 2026-05-12T20:57
My current video draft contains 191 mentions of "world model". That is more than 3 minutes of saying "world model" and nothing else 💀 → tweet link
Miscellaneous Tech
@FrameworkPuter · 2026-05-13T05:40
RT @QVHenkel: I started characterising the cooling solution for the Framework SDR. I printed an adapter for a germanium window to fit into… → tweet link
@badlogicgames · 2026-05-13T15:59
RT @eSaadster: released pi-cua-driver. Pi can now drive native macOS apps in the background using cua-driver: click/type/scroll without st… → tweet link
@jezell · 2026-05-12T20:38
RT @HoytEmerson: Yep...it's official. DuckDB now let's two clients connect and write to it. → tweet link