GPT-5.5 and GPT-5.5 Pro land in the API with 1M token context, built-in computer use, MCP, Skills, and medium reasoning effort default.
Key Takeaways
GPT-5.5 supports 1M token context, image input, structured outputs, function calling, prompt caching, Batch, and web search in one model.
Built-in computer use, hosted shell, apply patch, Skills, and MCP are native capabilities, eliminating external tooling setup.
GPT-5.5 Pro is Responses API only (including Batch), targeting high-compute workloads that benefit from additional inference budget.
Tool search defers large tool surfaces to runtime, reducing token overhead and preserving prompt cache performance and latency.
GPT-5.5 defaults to medium reasoning effort, reverting from GPT-5.1’s none default introduced November 2025.
Hacker News Comment Review
Benchmark results are inconsistent: one SQL benchmark scores it 25/25 matching Opus 4.7, while a WordPress/GravityForms benchmark ranks it last on both performance and value.
The tiered pricing cliff at 272K context drew skepticism: input doubles to $10/M above that threshold, making long-context workflows more expensive than Opus 4.7 without proportional efficiency gains.
OpenAI cited unresolved API safety requirements the day before launch with no public explanation, raising questions about what safeguard work was completed in under 24 hours.
Notable Comments
@wincy: GPT-5.5 returned an incomplete transaction stub requiring manual follow-up despite explicit rollback instructions, flagging persistent instruction-following gaps.
@robertwt7: GPT-5.5 paired with Codex delivers high-confidence results; contrasts with Opus 4.7’s CLAUDE.md instruction drift and hallucinations in daily use.
