Elevated Errors on API and ChatGPT

Original Article ↗ Hacker News Discussion ↗

Frequent outages and slowdowns in OpenAI’s API and ChatGPT, coinciding with new feature rollouts and recent corporate turmoil, are raising concerns about reliability among both paying users and developers building on the platform. Commenters debate whether rapid innovation justifies service instability, call for clearer SLAs or “stable vs. preview” tiers, and share alternatives such as Azure OpenAI, local models, and competitors like Phind. OpenAI staff acknowledge the incident, attribute it to scaling challenges with massive demand, and promise a postmortem.

Outage and Visible Symptoms

  • Users report OpenAI API and ChatGPT being down, with internal server errors, logouts, failed logins, and generic “problem with your request” messages.
  • Some see a new “ChatGPT Alpha” / “alpha models” dropdown, apparently related to experimental features, adding to confusion.
  • Status page uptime graphs for November are perceived as poor.

Reliability, Scaling, and Claimed Causes

  • Many assume the outage is tied to post–Dev Day traffic spikes and the rollout of free voice chat.
  • Someone speaking for OpenAI says it’s unrelated to both the voice rollout and the recent leadership drama, and attributes it to normal scaling challenges with a very large user base.
  • Others remain unconvinced, noting frequent outages and slowdowns since Dev Day, especially for GPT‑4 and gpt-4-1106-preview.

Impact on Paying Users and API Developers

  • Strong frustration that paid users and API workloads are disrupted by launches for free users.
  • Developers integrating the Assistants API and GPT‑4 models emphasize the need for graceful degradation and better reliability.
  • Questions arise about refunds and SLAs; another commenter notes the terms explicitly disclaim uninterrupted service.

Speed vs Stability, and Product Strategy

  • Split views: some value rapid iteration even with outages, others want “stable” vs “preview” endpoints and clearer “this is beta” messaging.
  • Comparisons are made to GitHub, Gmail, and other infrastructure-like services that users expect to be highly reliable.

Experiences with New Features and Models

  • Assistants API is generally liked, with issues around tool-calling accuracy and sometimes ignoring provided files.
  • Several find new preview models powerful but “insanely slow,” affecting usability.

Alternatives and Fallback Options

  • Suggested substitutes include Azure OpenAI (seen as more stable), Bing Chat, Phind, Claude, Pi, local models via Ollama (Mistral, Code Llama, OpenChat), Bedrock with Llama 2, and others.
  • Some report Phind and certain open models performing surprisingly well, especially for coding.

Dependence, Productivity, and Broader Reflections

  • Many realize how dependent they’ve become on ChatGPT for coding, writing, and note‑taking; outages significantly disrupt workflows.
  • One commenter with ADHD describes it as more helpful than medication; others worry about over‑reliance and loss of independent problem‑solving.

Tone: Humor and Anxiety

  • Thread mixes serious concern about OpenAI’s stability with jokes (HTTP error gags, sci‑fi heist scenarios, “404 CEO NOT FOUND”).
  • The incident amplifies existing worries about the company’s governance and long‑term reliability, even as users praise the product’s capabilities.