Executive Summary
The last 24 hours were dominated by OpenAI's launch of Codex in the ChatGPT mobile app and a personal finance preview for Pro users, signaling a strong push toward agentic, always-on AI assistants. Meanwhile, GPT-5.5 is under scrutiny as multiple developers report degraded performance and ballooning input token usage (up 2.5x). On the open-source front, NousResearch's Hermes Agent is rapidly gaining ground—claiming nearly 2x the daily token volume of OpenClaw just days after surpassing it—and Ollama rolled out native Codex app support alongside expanded NVIDIA Blackwell GPU capacity for GLM-5.1. Grok Build, xAI's agentic CLI coding tool, also entered early beta, intensifying competition in AI-powered development tools.
Key Events
- OpenAI launches Codex in the ChatGPT mobile app, enabling users to start work, review outputs, and steer executions from mobile—a major step toward universal agent usage. → link
- GPT-5.5 performance degradation reported: The Codex team acknowledged reports of GPT-5.5 performing worse for some users; separately, developers noted input token usage spiking 2.5x with declining output quality. → link → link
- ChatGPT personal finance preview launched for U.S. Pro users, allowing secure linking of financial accounts—positioning ChatGPT as a 24/7 personal agent. → link
- Ollama adds Codex app support and deploys more NVIDIA Blackwell GPUs for GLM-5.1 cloud inference, with one-command setup via
ollama launch codex-app. → link - Grok Build agentic CLI enters early beta for SuperGrok Heavy users, offering a TUI for coding and workflow automation. → link
- Hermes Agent token volume nearly 2x's OpenClaw just days after surpassing it, according to NousResearch's Teknium. → link
- MLX achieves full CUDA backend test pass, a milestone for Apple Silicon ML framework's cross-platform compatibility. → link
- X (Twitter) algorithm open-sourced with architecture diagram revealing a Grok-powered scoring component. → link
Analysis
Patterns: The competitive landscape in AI coding agents has reached a fever pitch. Within a single day, Codex expanded to mobile, Ollama added Codex support, Grok Build launched its beta, and Hermes Agent continued its explosive growth—all targeting the same developer "agentic coding" market. Simultaneously, the incumbent tools (GPT-5.5 via Codex, Claude Code) face user complaints about performance and speed, creating openings for alternatives.
Escalation: The arms race for token volume and developer mindshare between OpenClaw and Hermes is escalating rapidly. Meanwhile, infrastructure investment is accelerating—Ollama's Blackwell GPU expansion and the push toward local inference (DGX Spark, tinygrad, MLX) points to a bifurcating market: cloud-heavy vs. local-first AI deployment.
What to watch next: - Whether OpenAI addresses GPT-5.5 quality regression and input token inflation—this could erode trust in Codex if unresolved. - Grok Build's trajectory post-beta; early impressions are positive but it's SuperGrok-gated. - Continued momentum of "agent-native" tooling (CLIs, MCP integrations) as infrastructure shifts from chat assistants to autonomous agents. - ZK proof-based training data verification—if adopted, it could create a new data marketplace paradigm.
Tweet Feed
AI Models & Performance
@thdxr · 2026-05-15T05:03
i had a good several week run with gpt 5.5 today was completely miserable im so mad rn → tweet link
@thdxr · 2026-05-14T19:28
something is going wrong with gpt 5.5 caching
doesn't look like much on this chart but this it's now using 2.5x as many input tokens as a week ago and dropping → tweet link
@badlogicgames · 2026-05-15T17:44
RT @thsottiaux: Codex team is aware of reports of GPT-5.5 performing worse for some users and investigating. We don't have anything conclus… → tweet link
@levelsio · 2026-05-15T14:25
If Claude Code keeps being slow like this while I pay $200/mo (and they don't let me pay more)
They will essentially force me to leave to Codex and I don't want to
But it's soooooo slooooooooooooowwwww → tweet link
@ollama · 2026-05-15T03:37
We just added significantly more NVIDIA Blackwell GPUs to better serve GLM-5.1 model on Ollama's cloud.
We have been adding more GPUs daily for all the other models.
Claude Code: ollama launch claude --model glm-5.1:cloud Codex App: ollama launch codex-app Hermes Agent: ollama launch hermes --model glm-5.1:cloud Run the model: ollama run glm-5.1:cloud → tweet link
AI Products & Features
@sama · 2026-05-14T21:16
Codex in the ChatGPT mobile app! → tweet link
@gdb · 2026-05-14T21:15
You can now use Codex, wherever you have it running, from the ChatGPT app.
Huge step forward for universal usage of agents. → tweet link
@gdb · 2026-05-15T17:11
Understand and manage your personal finances in ChatGPT.
A further step towards ChatGPT becoming your personal agent, operating on your behalf 24/7, for helping you at home and work. → tweet link
@sama · 2026-05-15T18:35
i appreciate how seriously the team always takes these reports (even when the answer turns out to be 'i got used to the current level of magic and now i'd like more please') → tweet link
@jezell · 2026-05-15T05:27
RT @xai: An early beta of Grok Build, an agentic CLI for coding, building apps, and automating workflows is now available for SuperGrok Hea… → tweet link
@Ex0byt · 2026-05-14T22:29
An hour in, and my first impression of the Grok Build Beta TUI has me impressed → tweet link
@nummanali · 2026-05-15T15:32
Tempted to turn on Codex memories
For those that have tried it for >2 weeks, have you seen improved behaviour and recall from Codex?
And how has the token spend looked after turning it on, did you feel much difference in usage? → tweet link
Hermes Agent & NousResearch
@Teknium · 2026-05-15T03:17
I dont love to gloat but we are almost 2x'ing openclaw just 3 days after surpassing their daily token volume 🤗 → tweet link
@Teknium · 2026-05-14T21:38
RT @NousResearch: You can now power your Hermes Agent, if using OpenAI models, with codex as the runtime for the core tools that it offers,… → tweet link
@Teknium · 2026-05-14T22:23
We got your deepseek here, deepseek here!
Sign up for Nous Portal's free tier if you haven't to get some free DeepSeek V4 Flash in your Hermes Agent! → tweet link
@Teknium · 2026-05-14T21:40
I really like the Herm's UX with Nous Girl hanging out while you work in the TUI they built! → tweet link
@Teknium · 2026-05-14T21:38
Do way more with Hermes than just coding! → tweet link
@Teknium · 2026-05-15T15:21
RT @PurzBeats: My ComfyUI Template Integrity skill has now been merged into the official Hermes Agent repo. It makes Hermes a lot more comf… → tweet link
@Teknium · 2026-05-14T19:34
New Hermes Agent meetup alert!
This time in Arlington, TX! → tweet link
@ollama · 2026-05-15T17:07
RT @NVIDIA_AI_PC: Run @NousResearch's Hermes Agent fully locally on DGX Spark. 🚀
Our newest playbook shows you how to get set up via @Olla… → tweet link
Developer Tools & Agent Infrastructure
@ollama · 2026-05-15T01:38
Ollama now supports Codex app!
To try it, update to the latest Ollama 0.24, and run:
ollama launch codex-app
Select an open model to use with Codex app! → tweet link
@badlogicgames · 2026-05-15T17:53
sneak peak of the new https://t.co/oUoqqL9hAp no tools mode. you're gonna love it. you can try it with
pi -nbt. → tweet link
@badlogicgames · 2026-05-15T15:06
people of https://t.co/oUoqqL9hAp. i'm removing all tools from pi witbout replacement.
get creative. → tweet link
@steipete · 2026-05-15T05:49
CodexBar 0.26.0 is live
⚡ Kiro, Antigravity, OpenRouter, Kimi 🧭 calmer menus + keyboard nav 📊 better Codex/Claude limits and cost scoping 📦 named macOS assets, CLI + Homebrew fixes → tweet link
@steipete · 2026-05-15T06:47
This is a game changer. With codex autoreview and crabbox I can now go from issue to fix almost fully automated. (yes it does burn lots of tokens) → tweet link
@steipete · 2026-05-14T20:42
We've been working really hard on performance, reliability, security, and stability. Invented whole new automation flows with crabbox, automated video QA and are spending insane amounts of CPU cycles on CI.
It's a good release. → tweet link
@steipete · 2026-05-15T17:55
The latest CodexBar update renders API costs wayyyy nicer. → tweet link
@badlogicgames · 2026-05-15T12:48
RT @YoniBraslaver: Cloudflare's Code Mode post argued agents are more efficient with code than a menu of MCP tools.
We ran the experiment… → tweet link
@thdxr · 2026-05-15T12:43
i'm working on some sdk design for v2
what i keep coming back to is not fully understanding one segment of the usecase
every infra company is prepping for "millions of agents to run on us"
are you working on a product that needs this? tell me more → tweet link
@nummanali · 2026-05-14T20:28
A guide for Agent Native CLIs has been missing
Grateful to Jonathon and Notion team to have made this public
It cover both interactive and non interactive modes, as well differentiating human / agent interactions → tweet link
Open Source & Frameworks
@RydMike · 2026-05-15T18:54
RT @shiweidu: alien_signals 2.3.0 is out.
Adds effect() cleanup callbacks, improves effectScope nesting and cleanup ordering, and syncs wi… → tweet link
@FrameworkPuter · 2026-05-15T17:54
We'll be at Open Source Summit next week in Minneapolis if you want to check out the new Framework Laptop 13 Pro in person! We're at booth G/S16. → tweet link
@FrameworkPuter · 2026-05-14T21:31
Framework Wireless Touchpad Keyboard is by far our most waitlisted product! We may need to increase our plans for production capacity. → tweet link
@jezell · 2026-05-15T15:54
RT @zcbenz: We have achieved a milestone in MLX that all tests are passing in CUDA backend now. → tweet link
@nummanali · 2026-05-15T14:01
This is so cool X Algorithm open sourced
Checkout the architecture diagram in the readme, especially the Grok scorer → tweet link
@gospaceport · 2026-05-14T21:50
Open source CAD having an awesome moment → tweet link
@steipete · 2026-05-15T06:54
The latest release of OpenClaw is the first one that ships with our new TypeScript security hardening file-system lib. Previously, this was a grown mess of ad-hoc hardening which was hard to maintain, slow and inconsistent.
https://t.co/PXTXYFOIAQ increased some file ops by 10x. → tweet link
@gospaceport · 2026-05-15T17:51
These Open Source designs are pretty dope. If you ran gpus during the milk crate days this vibes like an updated modern super super crate. Design Files at the end tweet also! → tweet link
@tinygrad · 2026-05-15T04:56
RT @real_deep_ml: We just added both tinygrad and PyTorch problem collections. Curious to see which one people complete more → tweet link
Local AI & Hardware
@TheAhmadOsman · 2026-05-15T17:50
Mike is among the TOP 10 people working in Local AI that no one is paying attention to
His stuff is in-depth, well researched and reasoned
Follow this guy and congratulate him on winning AMD Developer contest for his Local AI work → tweet link
@TheAhmadOsman · 2026-05-15T02:03
Had a great time chatting with Alex yesterday
Dude is serious about what he's building he literally bought a DGX Spark / GB 10 on his way to show me some of Exo's upcoming cool features → tweet link
@sudoingX · 2026-05-15T07:41
RT @sudoingX: look anon, those of you who kept saying local AI is not there yet, who said open source can't compete, who said you need clou… → tweet link
@alexocheema · 2026-05-14T20:13
If you're considering dropping 20k on a 512GB M3 Ultra Mac Studio o eBay, consider 4 x M5 Max MacBooks with 6 TB5 cables instead.
Same price with 3x memory bandwidth, 9x FLOPS with all-to-all RDMA. → tweet link
@victormustar · 2026-05-15T10:13
RT @_lewtun: You can now have an AI researcher running on your laptop 24/7 for free!
Running Qwen3-35B-A3W with llama.cpp and a 4-bit qua… → tweet link
Programming & Developer Experience
@thdxr · 2026-05-15T18:34
branded types for absolute vs relative paths are gonna save my life → tweet link
@badlogicgames · 2026-05-15T18:20
effect wouldn't be necessary if TS had checked exceptions.
don't @ me. → tweet link
@badlogicgames · 2026-05-15T09:26
am i holding it wrong, or has VS Code auto-complete just given up being helpful? why does it take seconds to surface the args? → tweet link
@kunchenguid · 2026-05-15T16:22
built a dead simple chrome extension for my wife and myself. sharing here as free open source as usual
Simple Words
we find ourselves spending too much time wordsmithing our email replies, just to make sure we sound friendly and respectful
this chrome extension is really simple and quite literally a gpt wrapper that only does one thing - one click turning what you mean to what's ready to send
it runs on your own api key, subscription or local LLM. the extension itself is free and open source. → tweet link
@nummanali · 2026-05-15T12:36
This is your signal to use cmux
It uses native libghostty so your coding subscriptions are unaffected
Lawrence and team just get it, the experience is not opinionated i.e you use it to your workflows → tweet link
@thdxr · 2026-05-14T23:50
im not an honest person i squash my commits → tweet link
Research & Ideas
@nummanali · 2026-05-14T23:19
ZK Proofs for training data is really smart
In recent months smaller labs have gained traction by "cleaning" training data to be more high quality
If this happens, models would improve faster and their could be a data marketplace with royalties
Food for thought! → tweet link
@alexinexxx · 2026-05-14T20:18
today feels like a good day to dig into the AutoKernel paper → tweet link
@jsuarez · 2026-05-15T00:48
I am attempting to solve curriculum learning in RL in the next 2 weeks. Join me every day for 6-8 hours of livestreamed research, a 10 km run, heavy compounds, and a 1,000 calorie cut. Details + ground rules on day 1! → tweet link
@gospaceport · 2026-05-15T15:18
RT @snwy_me: i dug into this
in the transformer, all of the actual transformer parameters are zeroes?? so either they're hiding that or fo… → tweet link
@badlogicgames · 2026-05-15T17:44
The reason is simple: everybody is being inundated by the slop machine.
the future is glorious. the future also sucks. → tweet link