Shirt a Phobia

Sample Page

GPT-5.5 (Generative Pre-trained Transformer 5.5) is a large language model (LLM) released by OpenAI on April 23, 2026.[1][2] The model is also known by its codename “Spud”.[3]

OpenAI reported GPT-5.5 benchmark scores including 82.7% on Terminal-Bench 2.0, 51.7% on FrontierMath Tier 1–3, and 35.4% on FrontierMath Tier 4, while listing lower scores for Claude Opus 4.7 and Gemini 3.1 Pro on the same benchmarks.[4][5]

OpenAI said a recurring tendency in its models to mention goblins, gremlins, and other creatures began with GPT-5.1 and became noticeable in GPT-5.5’s Codex testing.[6] The company attributed the behavior to rewards used when training the “Nerdy” personality, which favored creature-word outputs and transferred beyond that personality during later training. OpenAI said it retired the Nerdy personality, removed the goblin-affine reward signal, filtered training data containing creature words, and added a developer-prompt instruction for GPT-5.5 in Codex.[7]

Release

GPT-5.5 Thinking and GPT-5.5 Pro were released on April 23, 2026, with neither being available to free-tier users. OpenAI withheld API access at launch, saying that API deployments required “different safeguards”; GPT-5.5 and GPT-5.5 Pro became available in the API on April 24, 2026.[5][8][9] GPT-5.5 Instant was released to free-tier users on May 5, 2026, replacing GPT-5.3 Instant as ChatGPT’s default model for all users.[10] On May 7, 2026, OpenAI announced GPT-5.5-Cyber, a limited-preview variant for vetted cybersecurity teams under its Trusted Access for Cyber program.[11]

System prompt

In April 2026, a GPT-5.5 developer-prompt instruction in the GitHub repository of Codex told it not to mention goblins if the user prompt did not require it.[12]

Reception

ZDNET praised GPT-5.5 for its polished answers and “[s]trong performance across writing, coding, and reasoning tasks”.[13] Compared to its predecessor, the website described GPT-5.5 as better, faster, and showing “improvements in agentic coding, conceptual clarity, scientific research ability, and accuracy during knowledge work”.[13]

The AI Security Institute reported that GPT-5.5 had a 71.4% (±8.0%, one standard error) average pass rate on its expert-level cyber tasks, compared with 68.6% (±8.7%) for Anthropic’s Claude Mythos Preview, and said GPT-5.5 “may be the strongest model we have tested” on that measure.[14]

References

  1. ^ Capoot, Ashley (April 23, 2026). “OpenAI announces GPT-5.5, its latest artificial intelligence model”. CNBC. Retrieved April 23, 2026.
  2. ^ Peters, Jay (April 23, 2026). “OpenAI says its new GPT-5.5 model is more efficient and better at coding”. The Verge. Retrieved April 23, 2026.
  3. ^ Mills, Madison; Fried, Ina (April 23, 2026). “OpenAI releases “Spud” GPT-5.5 model”. Axios. Retrieved April 23, 2026.
  4. ^ Franzen, Carl (April 23, 2026). “OpenAI’s GPT-5.5 is here, and it’s no potato: narrowly beats Anthropic’s Claude Mythos Preview on Terminal-Bench 2.0”. VentureBeat. Retrieved May 15, 2026.
  5. ^ a b “Introducing GPT-5.5”. OpenAI. April 23, 2026. Retrieved May 15, 2026.
  6. ^ McMahon, Liv (April 30, 2026). “OpenAI tells ChatGPT models to stop talking about goblins”. BBC. Retrieved May 13, 2026.
  7. ^ “Where the goblins came from”. OpenAI. April 29, 2026. Retrieved May 15, 2026.
  8. ^ “GPT-5.5 Model | OpenAI API”. developers.openai.com. Retrieved May 15, 2026.
  9. ^ “GPT-5.5 pro Model | OpenAI API”. developers.openai.com. Retrieved May 15, 2026.
  10. ^ “GPT-5.5 Instant: smarter, clearer, and more personalized”. OpenAI. May 5, 2026. Retrieved May 15, 2026.
  11. ^ “Scaling Trusted Access for Cyber with GPT-5.5 and GPT-5.5-Cyber”. OpenAI. May 7, 2026. Retrieved May 15, 2026.
  12. ^ Orland, Kyle (April 30, 2026). “OpenAI Codex system prompt includes explicit directive to “never talk about goblins”“. Ars Technica. Retrieved April 30, 2026.
  13. ^ a b “I put GPT-5.5 through a 10-round test: It scored 93/100, losing points only for exuberance”. ZDNET. Retrieved April 27, 2026.
  14. ^ “Our evaluation of OpenAI’s GPT-5.5 cyber capabilities | AISI Work”. AI Security Institute. April 30, 2026. Retrieved May 15, 2026.