Sam L

HUGE milestone! 😄 from a barebones Buildroot image to the first successful GPU inference test run with llama.cpp - qwen2.5 i have a massive smile on my face right now 😅

@benln Bangkok! nice 👌

Cafe Cursor is coming up in these cities: • LA - 7/15 • Philly - 7/16 • Lisbon - 7/17 • Zambia - 7/18 • Salvador - 7/20 • SF - 7/20 • Medellin - 7/22 • Rosario - 7/24 • Dallas - 7/26 • Cologne - 7/30 • Bali - 8/3 • Bangkok - 8/9 • Casablanca - 8/9 More being added

@elonmusk please delete... 😰

i used Grok Build in my C:\ the other day...

Computer Use in Codex is pretty impressive... it's currently learning how to use KruxOS in a VM and creating a skill that will be shareable so your agents can spin up VM's and know exactly how to navigate

@pwnies one of the most amazing designs i've ever seen Jacob but please fix this css 😅

What are we calling this interweave of classical painting styling and digital effects/shaders? Neoclassicore? Renderaissance? Baroquepunk? Whatever it is I'm in love with it.

what even is §

> pauses using Grok Build until clarified by @xai / @SpaceXAI

@pvncher hi eric! i'm using codex (chatgpt) in windows at the moment and my view looks completely different and can't see any diffs or anything. what am i doing wrong? i can see the file preview, but the agent keeps reverting to producing full diffs in-line in the conversation

One of the first big features I had a hand in shipping! Give it a try and click the pull request tab

@thsottiaux dam... didn't get a banked reset 🥲

Added a banked reset to 500k users of ChatGPT Work and Codex. What’s been happening: - Just released the ability to use a banked reset from web and mobile. Before today you could only do it in the desktop app - Had an issue where less than 10% of users who used a banked reset and it didn’t actually reset. This was during a 2 hour window. - It was hard to find all the exact users where it didn’t apply, so instead granted a banked reset to everyone who pressed the reset button in that 2 hour window. This also gives us the opportunity to validate the new infra ahead of tomorrow Tomorrow we will celebrate our 7M active users milestone and grant the first banked reset across all of our ChatGPT Work and Codex users. We will also release a bunch of updates to the desktop app, addressing a ton of your feedback. Talk soon

@wasjakehere @claudeai

@claudeai @samelldev Will take a reset as well.

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

@claudeai this is how we feel every time you post this

@brandon_galang Luna on Max effort scoring similar to Sol on Max but costing 5x less

don't fall for the bait to use gpt 5.6 luna because the benchmarks show its performance is "pareto optimal" it may perform similarly over a one shot assessment, but luna, terra, and sol are dramatically different in the output tokens and agent steps needed to finish the task sol is noticeably better at recall over long context when you're going back and forth with a model and trying to figure out what needs to get done, it'll benefit you to have the higher parameter count of sol even if it technically performs the same as terra or luna on benchmarks

Yesterday's replies recommended Luna for subagents. My Codex tests went the opposite way. Then I saw Tibo's post and moved the same benchmark to the Claude Code harness. Sol medium: 𝟰𝟲.𝟮% 𝗳𝗮𝘀𝘁𝗲𝗿 · 23.1% fewer output tokens Luna medium: 𝟮𝟳.𝟵% 𝗳𝗮𝘀𝘁𝗲𝗿 Back in Codex, Luna medium was the only faster config. It missed 3 real issues in the webhook review. Luna high and xhigh landed near Sol medium's quality and took longer. Same three tasks: → fix an async cache → implement an async DAG with concurrency, retries, and failure propagation → audit a payment webhook I used a fresh directory and session for every run, kept the prompts and hidden grading unchanged, and set a 360s timeout. I disabled MCP, skills, plugins, user and project settings, Chrome, subagents, and session persistence. 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝗵𝗮𝗿𝗻𝗲𝘀𝘀: completion · time · output tokens · tool calls · avg score 𝗦𝗼𝗹 𝗺𝗲𝗱𝗶𝘂𝗺 → 3/3 · 139.9s · 6,038 · 17 · 96.67 𝗟𝘂𝗻𝗮 𝗺𝗲𝗱𝗶𝘂𝗺 → 3/3 · 160.8s · 7,441 · 20 · 98.33 𝗟𝘂𝗻𝗮 𝗵𝗶𝗴𝗵 → 3/3 · 343.5s · 16,449 · 24 · 96.67 𝗟𝘂𝗻𝗮 𝘅𝗵𝗶𝗴𝗵 → 3/3 · 519.9s · 26,588 · 26 · 95.00¹ 𝗟𝘂𝗻𝗮 𝗺𝗮𝘅 → 3/3 · 591.4s · 30,924 · 24 · 96.67 Compared with yesterday's Codex harness: Sol medium → 46.2% faster · 23.1% fewer tokens · quality unchanged Luna medium → 27.9% faster · 21.4% fewer tokens · 81.67 → 98.33 Luna high → 1.3% faster · 1.9% more tokens · quality unchanged Luna xhigh → 8.1% slower · 24.6% more tokens · 96.67 → 95.00 Luna max → 13.5% faster · 2/3 → 3/3 · tokens not comparable because yesterday's run timed out ¹ xhigh's raw average was 88.33. The hidden grader's greedy matcher merged two separate issues. I checked the matches by hand: xhigh found all 6 seeded issues and added 3 extras. The corrected average is 95.00. I left the raw output untouched. 𝗪𝗲𝗯𝗵𝗼𝗼𝗸 𝗿𝗲𝘃𝗶𝗲𝘄: 𝟰𝟱 → 𝟵𝟱 Codex harness → missed 3 real issues · added 2 false positives Claude Code harness → found all 6 seeded issues · added 1 false positive Each task/config got one run, the same method as yesterday. One run leaves room for sampling variance. The Sol medium speedup reproduced: 142.1s alone, 139.9s in the five-config parallel run, about 1.5% apart. I don't trust Claude Code's displayed dollar cost for Luna. It appears to estimate custom-model cost using a generic price, and it doesn't expose subscription quota conversion. I used GPT-5.6 Sol Extra High as the hidden grader, so I treat the quality scores as directional. Max exited this time after timing out yesterday, while xhigh slowed down under Claude Code. The higher-effort configs still look unstable. If you don't mind the extra setup, I'd try: → 𝗴𝗲𝗻𝗲𝗿𝗮𝗹 𝗱𝗲𝗳𝗮𝘂𝗹𝘁: Claude Code harness + Sol medium → 𝗹𝗼𝘄𝗲𝗿 𝗔𝗣𝗜 𝗰𝗼𝘀𝘁 + complex-task quality: Claude Code harness + Luna medium → 𝘀𝗸𝗶𝗽 𝗮𝘀 𝗱𝗲𝗳𝗮𝘂𝗹𝘁𝘀: Luna high / xhigh / max

@Voxyz_ai great tests Vox

@Blackwellboy ah needed one the other week, sorted now though i'll ask around for you mate 👍

@samelldev selling one yes XD

selling a kidney to fund more dgx sparks, anyone

Fable or cancel.