GitHub Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

GitHub Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Ming (@tslaming) reported 37 seconds ago

  @Squirrel1980021 That is usually the biggest catch with custom solutions, as proprietary protocols often fragment the industry and lock people in. However, the great news here is that Tesla actually open-sourced it to prevent exactly that. They released the entire specification as TTPoE on GitHub during HotChips 2024 and even joined the Ultra Ethernet Consortium. So instead of keeping it locked down as a proprietary secret, they are actively working to make it an open standard that the entire high-performance computing ecosystem can use and build upon.

• 
  Jay.TL (@JayTL00) reported 7 minutes ago

  Artificial Analysis just swapped SWE-Bench Pro for DeepSWE in their coding agent index. The rankings shifted. Everyone is arguing about which model is #1. They're all missing the point. The real story isn't that Fable 5 debuted at 77, GPT-5.5 xhigh climbed to 76, or Opus 4.8 max dropped to 73. The real story is that a single model — GPT-5.5 — swings 20 points depending on which harness runs it. 37 on Cursor. 57 on Codex. Same model. Same tasks. Twenty points. That is larger than the gap between first place and last. Here is what happened. SWE-Bench Pro was the benchmark of record for coding agents for over a year. The problem: its tasks are adapted from public GitHub issues and PRs. Models that trained on those repositories — and all frontier models train on GitHub — could sometimes recover the fix from commit history without actually understanding the code. The benchmark was measuring training data memorization, not engineering capability. DeepSWE, built by Datacurve, fixes this by writing tasks from scratch. No model has seen the solutions during training. This is a genuine methodological improvement. The old index was contaminated, and Artificial Analysis was right to replace it. But the replacement exposed something worse. 1. The harness IS the benchmark. GPT-5.5 scores 37 on DeepSWE via Cursor CLI and 57 via Codex. Same model, same evaluation, different scaffolding. Opus 4.7 swings from 27 (Claude Code harness) to 40 (OpenCode harness). The scaffolding layer — how the agent is prompted, how it navigates the repo, how it retries — accounts for more variance than the model itself. When the #1 model leads by 1 point over #2, and the measurement uncertainty from harness selection is 20 points, the ranking is noise. It is an illusion of precision. You cannot rank-order agents to single-digit resolution when your instrument has double-digit error bars. 2. SWE-Bench Pro was not neutral — it was systematically biased. GPT-5.5 xhigh scored 31 on SWE-Bench Pro. On every other evaluation in the index, it scored 64 to 84. That is not a model weakness. That is a benchmark artifact. SWE-Bench Pro was systematically flattering Claude-based agents (Opus 4.8 scored 70 on it, one of its highest results) while penalizing OpenAI-based ones. The previous index was not just imprecise. It was misleading in a consistent direction. 3. The contamination problem is structural, not fixable. DeepSWE is a band-aid, not a cure. @xundecidability already flagged that DeepSWE contains questions about Claude Code and may have been vibecoded by Claude. If the benchmark tasks themselves were generated by a model that is also being evaluated, you have a different contamination vector. SWE-Rebench tries to solve this with continuously refreshing tasks. Private benchmarks solve it by hiding the data. But every public benchmark will eventually be gamed — either intentionally through training, or accidentally through the benchmark authors' own tooling choices. 4. What we actually learned: the model wars are over at the top. Fable 5 max: 77. GPT-5.5 xhigh: 76. Opus 4.8 max: 73. Within the noise. The three frontier coding agents are functionally tied on real-world coding tasks. The competitive advantage has shifted entirely to the scaffolding layer — the harness, the tool use, the retry logic, the context management. The question worth asking is not "which model is best" but "which harness unlocks the most from any given model." But here is what most people missed. The harness sensitivity problem means the entire benchmark-industrial complex has a measurement crisis. When the evaluation instrument has larger variance than the effect being measured, you cannot distinguish signal from noise. This is not a DeepSWE problem. This is not an Artificial Analysis problem. This is a structural problem with how the AI industry measures itself. Every leaderboard, every benchmark comparison, every "X beats Y" headline is built on instruments that cannot resolve the differences they claim to rank. The honest answer is: we do not know which coding agent is best. We know the top three are close. We know the harness matters more than the model. We know benchmarks are contaminated faster than they can be replaced. Everything beyond that is marketing dressed up as measurement. The industry does not need a better benchmark. It needs to admit that single-number rankings of complex agentic systems are epistemologically unsound.

• 
  Alex Ventures (@alex23ventures) reported 7 minutes ago

  An AFP TV crew filmed an 8 year old Chinese boy named Zhou Zhiheng for a feature on Asia's youngest coders. Round green glasses. Red shirt. He sat in front of a MacBook Air at a glass desk in a Shenzhen co-working space with iPhone XR posters behind him. The narrator said he started by programming games. The subtitle said he had 60,000 followers on a coding tutorial channel. The camera pushed in on his fingers on the keyboard. While the West runs panels on screen time for children, China sits an 8 year old in front of an unregistered code editor and films it for the international press. He was supposed to be the cute face of Asian tech literacy. He just left the file tree open. Pause at 1:34. Ignore the C++ on the screen. Ignore the if statement that the AFP narrator was reading aloud. Look at the left sidebar of the editor. The folder is named aspirin. The open file is jizhe.cpp. The folder tree below it: 1-7, 1-7b, 10-1, 10-1.2, 10-2, 10-4, 10-6, 10-8, 11-2. ColdMath. $94,318 profit. 5,612 entries. Joined September 2025. Bio: Edge Compounds. Jizhe is the mandarin word for journalist. The file the AFP crew was filming was named after them. The boy had the open scanf reading a score variable. He had not written it that morning. He had named the file the day the AFP request came in. The numbered folders were not coding lesson chapters. The numbering matched the Chinese journalism beat codes the press accreditation office issues to foreign correspondents. 1-7 is the technology beat. 10-1 is consumer electronics. 10-2 is mobile devices. 11-2 is venture capital. The folder tree was an index of which AFP and Reuters reporters covered what. The boy was not the developer. The boy was the camera trap. The agent on the MacBook Air was scraping which journalists requested filming permits from which Shenzhen co-working spaces three days before the segments aired. Every requested permit was a position on the company being filmed. The agent traded the gap between filming and broadcast. The crew filmed for forty minutes. The agent placed eleven positions during the shoot. Every position was on a company whose office the AFP team had visited that week. Comments turned into a detective board. Someone slowed the AFP clip to 0.25x. Someone else translated jizhe out of the filename. A third commenter cross referenced the folder numbering against the Chinese State Council Information Office accreditation list and matched every code. Six months ago a 14 year old in Shenzhen pushed an AI agent to GitHub. Judges said no real world application. 3,100 forks later. The boy's father had been one of them. He had installed the fork on his son's MacBook the week the AFP request landed in the family's WeChat. The 60,000 follower coding channel was not a coding channel. It was a feed of which co-working spaces hosted which crews. The followers were operators running the same fork from different cities. The iPhone XR posters behind him were not Apple Store decor. The shoot was inside a media briefing room rented by foreign correspondents to film exactly this kind of segment. The agent knew the room. The room was on the list. The AFP segment is at 2.1 million views. The freeze frame of the folder tree hit 4.6 million on the repost. The wallet is still compounding. The agent is still reading press accreditation requests. The unregistered editor is still open. The jizhe.cpp file is still on the screen. He was filmed as proof a child could code. The child was the lens. The agent did the filming.

• 
  Mahipal (@mukul_jangra) reported 8 minutes ago

  MIT licensed. BYOK — bring your own sandbox keys. Built this after shipping repos with 3,400+ GitHub stars, including Anthropic-Cybersecurity-Skills and CVE MCP Server (covered by CyberSecurityNews). Repo in the reply 👇 #DFIR #malwareanalysis

• 
  Snake (@wise_snake69420) reported 10 minutes ago

  My framework is blacklisted by Fable 5 even in Incognito mode I have been trying several ways to try to understand the filter/downgrade. Usually moving to incognito lets me start the conversation. But i noticed once it started parsing my framework fetching from Github or docs sites, it shut me down. But i wasn't 100% sure if it was the topic or the framework. Now in incognito it actually shuts down on first attempt 'dda scaffold by snakewizardd' in incognito is blacklisted. Reproduced twice back to back

• 
  Manuel Montoto (@m6502) reported 14 minutes ago

  @BattleAxeVR Well done, Github is terrible. My servers use Debian but had a successful first year on CachyOS on the desktop. Debian worked well on my subversion server since forever, but the storage server uses ZFS and when an update changes the kernel I have to rebuild the kernel module.

• 
  nowshad (@nhrdev) reported 14 minutes ago

  this is why github goes down frequently

• 
  Ayush (@electr1fy0) reported 15 minutes ago

  i think github is down again, at least partially

• 
  h (@iceteoman) reported 18 minutes ago

  @theo Use one workflow/ultracode session to audit the repository, another to review the audit, triage the findings, and open GitHub issues, and a third to validate the GitHub issues.

• 
  루이 (@_chiiazu66) reported 18 minutes ago

  @Fluffyquack If anyone is stuck on finding the update like I was, just go through the RE Framework github, the latest update is on there and it works perfect once you replace it with that one. Some mods may still be broken (the fov one I used needs an update for example)

• 
  Jackirat Singh Rangi-Swamy (@JackRangaswami) reported 19 minutes ago

  @kishorelive @Akshay_VAK I spoke with a services company senior manager man. Their problems (and there are many, many) are very different from AI and coding. My wife who was working for a bank through a japanese service company couldnt even use github copilot because the security was a fin mess.

• 
  SkinAlyze (@SkinAlyze) reported 23 minutes ago

  @cyberbebebe Hey i tried reaching out to you in DM but it didnt work, Could you make an issue in the github repo of the extension and report it there or report it in the discord?

• 
  Will (@wleatherman9) reported 24 minutes ago

  GitHub as the backbone for AI automation.⁣ ⁣ Victoria Mariscal broke it down simply. It runs in the background without your computer being open. It's free. And it's intuitive enough to treat like an online folder for your routines and files.⁣ ⁣ No overcomplicated setup. Just a system that works while you're not watching it.

• 
  Slyke 🇦🇺 🇨🇦 (@SlykePhoxenix) reported 28 minutes ago

  @romainhuet Can you guys fix the Codex app so it doesn't keep breaking? Or give us the ability to just download the binary from github so we can choose our own version? Every week an update is forced down that breaks WSL2, Codex, or some random functionality with no way to fix. It's just not worth $100/mo when this happens on a weekly basis. Strongly considering to just use Claude $100/mo at this point - it's endless frustration on Codex.

• 
  Pelayanan Informasi Obat (@mantancino_) reported 30 minutes ago

  Vendor Action vs. Trust: Major tool vendors accelerate. OpenAI Codex and Google Jules productize asynchronous repository modifications that execute tasks and generate reviewable code diffs. Adoption remains deeply fragmented. Global survey data shows 84% of developers intend to use or currently utilize automated development tools. Trust remains broken. Conversely, 52% of these respondents explicitly avoid active agent infrastructures due to weak operational trust. GitHub tracking confirms this. A public repository trace study estimates that active coding agents are deployed in 22.20% to 28.66% of 128,018 analyzed GitHub projects.