OpenAI releases GPT-5.5 and GPT-5.5 Pro in the API

Original Article ↗ Hacker News Discussion ↗

Release timing & rollout

  • Some speculate the accelerated release was a response to DeepSeek; others think it was just final flag checks or that DeepSeek v4 is underwhelming.
  • Confusion over “safeguards and security requirements” mentioned the day before and how those could be resolved so quickly.
  • Rollout lagged for some enterprise and third‑party tools; a few users still saw only 5.4 initially.

Use cases & perceived value

  • Pro/expensive models are used for high‑value, infrequent tasks where cost is negligible compared with outcome (e.g., legal docs, ToS/PP drafting).
  • Some feel the marginal quality gain justifies the price; others don’t see meaningful improvements over cheaper models.

Safety, safeguards & liability

  • Strong disagreement over safety filters.
    • One side: filters are “counter‑productive,” harm access to medical and practical knowledge, and mainly shift liability away from providers.
    • Other side: hallucinations and mistranslations in contexts like medicine create serious risk; providers want to avoid PR/legal fallout.
  • Debate over real‑world alternatives for translation/diagnosis (professional interpreters vs AI vs “no help at all”).

Knowledge cutoff confusion

  • API docs list Dec 2025, but the model reports June 2024 in its own system prompt.
  • Several note model‑reported cutoffs have always been unreliable; practical testing suggests knowledge into early 2025.
  • Hypotheses: training data contamination, intentional older cutoff in prompts to encourage tool use; overall “unclear.”

Model quality & behavior

  • Mixed coding anecdotes: some see 5.5 as “shockingly good” and solving hard problems quickly; others see laziness (omitting obvious code) or no real gains over recent generations.
  • Long‑running automated coding workflows (hundreds of millions of tokens) reported as feasible and high quality by some; others are skeptical and expect “AI slop.”

Benchmarks & comparisons

  • Some benchmarks show 5.5 near or above top models (e.g., perfect SQL benchmark score, strong coding‑reasoning results).
  • Other user‑made benchmarks (e.g., WordPress plugin task) rank it poorly on both quality and value, with surprising underperformance versus some competitors. Methodology is debated.

Pricing, ecosystem & ethics

  • 5.5 (and especially 5.5 Pro) is significantly more expensive than 5.4 and Opus 4.7; concern that “subsidized AI” is ending and providers are clawing back margin.
  • Complaints about GitHub Copilot tiers and high multipliers; some predict migration to cheaper Chinese providers.
  • Ethical worries about financially supporting OpenAI, including references to alleged government surveillance contracts and concerns about astroturfing in online discussions.
  • Some report strict refusals on topics like benign SARS‑CoV‑2 analysis as evidence of over‑cautious safety policies.