Ask HN: If you've used GPT-4-Turbo and Claude Opus, which do you prefer?

Hacker News Discussion ↗

Developers and power users comparing OpenAI’s GPT‑4‑Turbo with Anthropic’s Claude 3 Opus report a clear tilt toward Claude for coding, summarization, and email-style writing, citing its more concise output, less “corporate” tone, and fewer signs of obvious AI authorship. GPT‑4, especially non‑Turbo variants, is still seen as stronger for complex logical reasoning, interpreting research papers, language learning, and workflows that rely on built‑in tools like Python execution, but many complain that recent GPT‑4 models feel “lazier” or degraded. Practical factors such as Claude’s limited availability in Europe, differing UI quality, and trust in each company’s direction also play a significant role in which model people choose.

Overall Usage Patterns

  • Many use both models and pick per task; some run GPT‑4, Claude Opus, and Gemini in parallel, then continue the conversation with whichever “gets” the task best.
  • Common use cases: coding/debugging, business planning, language learning, email and prose writing, summarization, diagram generation (e.g., Mermaid), and strategy/brainstorming.

Perceived Strengths of Claude 3 Opus

  • Coding: frequently described as “ahead” of GPT‑4, more to-the-point, better at understanding edits within existing code and making concise diffs. Some say it can be an integral part of an engineering workflow.
  • Style: less “corporate speak,” more concise, and less obviously AI-written; often preferred for emails and general writing.
  • Focus: tends to stay on topic, amend answers incrementally, and handle complex, multi-step instructions better in some users’ tests.
  • Multilingual: praised for certain low-resource languages and dialects.
  • Free Sonnet tier is considered clearly better than GPT‑3.5 by several; Opus seen as a noticeable upgrade over Sonnet.

Perceived Strengths of GPT‑4 / GPT‑4‑Turbo

  • Logic and papers: some report GPT‑4 as superior for complex logical reasoning and interpreting academic papers.
  • Language learning: GPT‑4‑Turbo favored for grammar/syntax explanations, rule generalization, mnemonics, and examples.
  • Reliability: a few find GPT‑4 less prone to confidently wrong answers on tooling/how‑to questions, especially when guided by a strong system prompt that discourages hallucinations.
  • Tools: integrated code interpreter and broader ecosystem (apps, plugins) are strong draws.

Style, Laziness, and Prompt Tuning

  • Multiple comments say GPT‑4 responses have become lazier or more generic over time, possibly due to cost or alignment tuning; it sometimes needs a second prompt to “really try.”
  • Claude is often perceived as less lazy and more willing to do full work.
  • Custom system prompts on GPT‑4 can reduce verbosity and “corporate” tone; Claude’s chat UI offers less system-prompt control.

Quality Disagreements and Hallucinations

  • Some find Claude markedly better for optimization/algorithms; others find it worse for bug fixing or factual “how do I do X in tool Y” questions, with stubborn hallucinations.
  • GPT‑4‑Turbo is divisive: some say it’s near original GPT‑4 quality; others call it “basically unusable,” especially for code (producing high-level steps instead of concrete implementations).
  • Benchmark reports in the thread conflict: one person’s internal benchmarks have several older GPT‑4 variants beating Claude Opus; another informal test set had Claude solving all problems GPT‑4 failed.

UX, Access, and Policy Friction

  • ChatGPT UI is seen as more polished (editing, canceling) but also buggy for some; Claude’s interface can crash or is missing advanced features.
  • Claude is not officially available in much of Europe; users resort to Poe, Kagi, proxies, or US payment workarounds.
  • Some users favor Claude for non-technical reasons: greater trust in its corporate direction and discomfort with previous OpenAI governance issues.