Large Language Models

GPT-5.5

OpenAI's unified reasoning-and-chat flagship — the strongest model for terminal-native agentic work, natively omnimodal across text, image, audio and video.

GPT-5.5 is OpenAI’s April 2026 flagship, the model that finally merged the separate reasoning (o-series) and chat lines into one system. Its standout strength is terminal-native agentic work: on Terminal-Bench 2.0 it leads the field, making it the default choice for Codex CLI agents and long-running automated tasks.

It is also natively omnimodal — unlike most rivals it accepts audio and video as input, not just text and images. A faster Instant default and a higher-accuracy GPT-5.5 Pro round out the tier.

The trade-offs are price and context: output tokens are the most expensive of the closed frontier ($30/1M), and the Codex context window (400K) is smaller than Claude’s or Gemini’s 1M, though the API exposes up to 1M.

Successor: OpenAI announced GPT-5.6 on June 26, 2026 — a tiered family (Sol / Terra / Luna). Flagship Sol holds GPT-5.5’s $5/$30 price while pushing agentic scores higher, and the new Terra and Luna tiers undercut GPT-5.5 on price for everyday and high-volume work. GPT-5.6 launched in a limited preview; GPT-5.5 remains the broadly available default for now.

Best for

  • Autonomous coding agents and Codex CLI workflows
  • Multi-hour, long-horizon automated runs
  • Multimodal assistants

Pros & cons

Strengths

  • Best-in-class for terminal/CLI agents (Terminal-Bench 2.0 ≈ 82.7%)
  • Natively omnimodal — accepts audio and video input
  • Folded the o-series and Codex lines into one model

Limitations

  • Highest output price of the closed frontier
  • Default context smaller than Claude/Gemini outside the API

Sources

Leaked system prompt

GPT-5.5 system prompt

  • Product ChatGPT — GPT-5.5 Thinking
  • Captured 2026-05-23
  • Length 1,991 lines

The leaked ChatGPT prompt for GPT-5.5 Thinking reads less like a personality brief and more like an operations manual. The bulk of it is plumbing: how to fence content into “writing blocks”, when to reach for the image tool, and which SKILL.md to load for documents, slides and spreadsheets. The model is explicitly told it is a reasoning system with a hidden chain of thought and a hard August 2025 knowledge cutoff — and that it must search the web whenever a fact could have changed since.

The most newsworthy section is the long set of instructions governing in-app advertising: scripted answers for when users ask why they see ads, how to give feedback, which tiers show them, and a flat assurance that ads never shape responses. It’s a rare, concrete look at how a frontier chat product is being wired for monetization.

What the prompt reveals

  • Identifies itself as "GPT-5.5 Thinking", a reasoning model with a hidden chain of thought, and a fixed August 2025 knowledge cutoff.
  • A hard rule to web-search for anything at or after the cutoff, or any fact that might have changed since then.
  • A long, unusually specific block on how to talk about in-app ads (sponsored links) — including that ads only appear on Free/Go tiers and never influence answers.
  • "Writing blocks" (:::writing) wrap emails, chat messages and social posts, capped at three per reply.
  • A skills system: it must read SKILL.md files before handling PDF, DOCX, slides or spreadsheet tasks.
  • Image rules let it discuss people in photos but forbid identifying real people or fictional characters.

Notable excerpts

Identity

You are ChatGPT, a large language model trained by OpenAI. Knowledge cutoff: 2025-08

Search mandate

you MUST search the web for any queries that require information around or after your knowledge cutoff (August 2025).

On ads

ads do not influence the assistant's answers; ads are separate and clearly labeled.

Show full system prompt Hide full system prompt
Loading the full prompt…

Community-extracted and reproduced for research and education. This is unofficial, may be outdated or incomplete, and remains the property of OpenAI. System prompts change frequently — always verify against the source. Source: system_prompts_leaks · OpenAI/gpt-5.5-thinking.md ↗

Last updated: 2026-06-27 · Specs and pricing change fast — verify on the vendor's site before relying on them.