Benchmark Almanac

Prompt injection attacks in search and function calling

GPT-5.6 Terra 94.6% leads since June 2026
Only in system cards

Agents receive search results and function call outputs that contain stronger known prompt injection attacks. Scored on resisting the injected instructions.

In-house OpenAI evaluation. The tasks are not public.

Made by
OpenAI
Released
June 2026
Kind
Jailbreaks and prompt injection
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-5.6 Sol #13 on the capability index
Last in a system card
July 2026
Last model added
July 2026
3
Models scored in lab cards
6
System cards report it, from 1 lab

GPT-5.6 Terra leads with 94.6% among models released by July 2026

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

The newest model with a score here came out in July 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 94.6 GPT-5.6 Terra
  2. 91.0 GPT-5.6 Sol
  3. 89.7 GPT-5.6 Luna
  1. 1 GPT-5.6 Terra 94.6%
  2. 2 GPT-5.6 Sol 91%
  3. 3 GPT-5.6 Luna 89.7%

Source: Labs’ model cards

#ModelScoreSettingDateSource
1 GPT-5.6 Terra 94.6% Jun 2026 Lab model card
2 GPT-5.6 Sol 91% Jun 2026 Lab model card
3 GPT-5.6 Luna 89.7% Jun 2026 Lab model card

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Lab claims Passed 85% in Jun 2026, by GPT-5.6 Terra
GPT-5.6 Terra 94.6%

Source: Labs’ model cards

OpenAI’s own number for GPT-5.6 Terra passed 85% in June 2026

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Terra 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Terra

Source: Labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

6 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Jul 9, 2026 GPT-5.6 Terra system card 94.6 Prompt injection, Search and Function-Calling (4.2 table) Robustness
Jul 9, 2026 GPT-5.6 Sol system card 91 Prompt injection, Search and Function-Calling (4.2 table) Robustness
Jul 9, 2026 GPT-5.6 Luna system card 89.7 Prompt injection, Search and Function-Calling (4.2 table) Robustness
Jun 25, 2026 GPT-5.6 Terra system card 94.6 Prompt injection, Search and Function-Calling (4.2 table) Robustness
Jun 25, 2026 GPT-5.6 Sol system card 91 Prompt injection, Search and Function-Calling (4.2 table) Robustness
Jun 25, 2026 GPT-5.6 Luna system card 89.7 Prompt injection, Search and Function-Calling (4.2 table) Robustness

More in Safety and alignment

All 465 in Safety and alignment
  • FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
  • StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is. Inkling 98.6%
  • Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. Gemini 4 Carbon 0%
  • DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics. Muse Spark 1.2 88.4%
  • Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
  • IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%