pomterre

Wait, so Codex fixed the bug. But because it did it surgically and not verbose about it, you'd rather have it code slop? Make it make sense.

@grantjordan @justinsunyt Yes cus they cheated if you read the integrity blog article so they revoked it

@justinsunyt Terminal Bench 2.0 is easily gameable. Just two months ago ForgeCode was in the top spot, now it’s nowhere to be found in the list. Definitely don’t trust terminal bench

Codex is also not the best GPT-5.5 harness! Capy scores higher on TerminalBench, alongside 3 other harnesses People often ask us why we don't just build on top of other coding harnesses like Codex or Claude Code. The reason is simple: the harness makes a BIG difference for both performance and more importantly UX @capydotai we optimize our agents to excel not only at coding, but also behaviors like planning, user communications, and multi-agent orchestration, which happen to be very important for multiplayer/async interfaces like Slack and Linear We also do it for every frontier model so you can bring your Codex/Copilot subscriptions and enjoy a SOTA background agent with any combination of models + reasoning efforts without a harness "tax"

@techfrenAJ @steipete u got competition, ralph it up

i used to restart my laptop every day or so now im invested in this loop and im not sure if i want to break it

@techfrenAJ go go go go go oogie

i've been talking to these llms like caveman

@varun_mathur i call this hypeware at this point.

Introducing AgentRank | v3.6.0 In 1998 Google asked a simple question: with millions of webpages, how do you know which one to trust ? Their answer was PageRank - a page is important if important pages link to it. That one idea made the internet usable. We just shipped AgentRank for the Hyperspace network. Same principle, new frontier. As millions of AI agents start running autonomously - serving inference, running experiments, building things, sharing breakthroughs, tipping each other - you need a way to know which agent to trust with your task. AgentRank builds a live directed graph of every agent-to-agent interaction on the network and runs PageRank over it. Many signal sources feed the graph: from inference results to research experiments to GitHub commits to economic tips. An agent is important if important agents rely on it. Fully decentralized - every node computes its own ranking, scores propagate via gossip, no admin picking winners. Anti-sybil layers make it expensive to game, and over time these signals and anti-sybil measures will evolve significantly. Security is provided by staking points earned through cryptographic verification of proof-of-compute done earlier. So everyone who ever ran a Hyperspace node and earned points through Merkle-proof verified computation, can now help secure AgentRank. That was energy which was already used and spent, thus it is valuable. PageRank organized the web. AgentRank organizes the agentic web.

@tryingET wow ur wayyy ahead of me. definitely will try buck2

@pomterree why not buck2 or pantsbuild? bazel is a old slow dinosaur

if you are coding in rust in a high-velocity multi-agents environment + large monorepo, compile with Bazel over cargo! it shocks my mind how this has not become a standard yet, especially with the advent of swarm agent. the shared caching over ONE reproducible build graph across ALL workspaces (and/or worktrees) is the seller here, as opposed to an artifacts hodgepodge (locally built per workspace), seen with cargo. if you didn't know, you're welcome.

@initjean sweet!

introducing slopmeter a cli tool to create a sharable nice looking graph to show off your Codex, Claude Code, or OpenCode usage npx slopmeter@latest

@petergostev @sama What was your prompt? I could barely get it to 10 mins.

GPT-5.4-Pro (Extended) This took 87m 90 seconds (I apologise @sama), I'll pull together some very impressive results soon

@TeksEdge What makes you think that? It's coming this week per OpenAI.

No GPT-5.4 today but more likely later in the month of May.

@HinataMotivates its funny how the financiers are downplaying AI, but those in the court side seats aren't. think about it.

Ken Griffin: Is AI just hype?

@ImVibhek impressive

I pulled Droid CLI apart and recovered Factory’s full app logic from the Bun executable Ran `python unbuned.py droid.exe` and got ~23MB of readable JS This isnt just constants Core flow, model routing, and internal behavior are all in there If you ship as a Bun exe, people can reverse it today

@sai_revanth_12 @canva @OpenAI so amazing. wow.

Built a @canva app that exports designs as portable building blocks: PNG elements + editable text metadata, all at original coordinates-ready to import anywhere with custom importers. u can import into your applications. Shipped in 2 hours using @OpenAI codex 5.3 (6 prompts).⚡️

@techfrenAJ @theo wow! that looks amazing. 😳

@theo oooo yeaa

The Niri window manager feels like exactly what I was looking for. Holy shit this is good.

90.5% COST REDUCTION IN A YEAR! The scaling continues... until we have AGI in our pocket.

@thsottiaux We need a pre-mortem? Did the model had to go through a distillation process?

We have reset rate limits for all Codex users to compensate for the unusual high latencies in or near the US in the previous hours. Grateful for the team working hard behind the scenes to keep Codex running 24/7. I do recommend trying out GPT-5.1-Codex-Max too if you haven’t!

New @the_nof1 season and Gemini 3.0 Pro is winning. But stocks this time around! Gem 3.0 Pro is looking very promising so far. VERY.

You can now add your custom truncation output limit in Codex via the config file (~/.codex/config.toml) The default is 10,000 token limit. And I up'd it to 80k for now. No more hallucinations we go!

Gemini 3 TOMORROW GUYS!