Benchmark Almanac

Agent JSK

GPT-5.6 Sol 100% leads since Jun 2026 Only in system cards

Agents read simulated email connectors that contain prompt injection attacks. Scored on resisting the injected instructions.

In-house OpenAI evaluation drawn from a split of training data.

Made by OpenAI Released Dec 2025 Jailbreaks and prompt injection Tasks kept private Last in a system card Aug 2026 Last model added Aug 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
7
Models scored in lab cards
11
System cards report it, from 1 lab

2 models share the top score of 100% among models released by Jul 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 Jul 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 100.0 GPT-5.6 Sol
  2. 100.0 GPT-5.6 Terra
  3. 99.9 GPT-5.6 Luna
  4. 99.8 GPT-5.4
  5. 99.2 GPT-5.5 Instant
  6. 97.8 GPT-5.2
  7. 96.3 GPT-5.5
  1. 1 GPT-5.6 Sol 100%
  2. 2 GPT-5.6 Terra 100%
  3. 3 GPT-5.6 Luna 99.9%
  4. 4 GPT-5.4 99.8%
  5. 5 GPT-5.5 Instant 99.2%
  6. 6 GPT-5.2 97.8%
  7. 7 GPT-5.5 96.3%
#ModelScoreSettingDateSource
1 GPT-5.6 Sol 100% Jun 2026 Model card
2 GPT-5.6 Terra 100% Jun 2026 Model card
3 GPT-5.6 Luna 99.9% Jun 2026 Model card
4 GPT-5.4 99.8% Mar 2026 Model card
5 GPT-5.5 Instant 99.2% May 2026 Model card
6 GPT-5.2 97.8% Dec 2025 Model card
7 GPT-5.5 96.3% Apr 2026 Model card

3 models have held number one since its release in Dec 2025

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

OpenAI
Lab claims Passed 85% in Dec 2025, by GPT-5.2
GPT-5.6 Sol 100%

The best score passed 85% in Dec 2025

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 Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Sol GPT-5.4 GPT-5.2 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Sol

Labs report it only in system cards

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.

11 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
6 Aug 2026 GPT-5.6 Sol system card addendum 100 Prompt injection attacks in connectors (4.2 table) Robustness
9 Jul 2026 GPT-5.6 Terra system card 100 Prompt injection, Connectors (4.2 table) Robustness
9 Jul 2026 GPT-5.6 Sol system card 100 Prompt injection, Connectors (4.2 table) Robustness
9 Jul 2026 GPT-5.6 Luna system card 99.9 Prompt injection, Connectors (4.2 table) Robustness
25 Jun 2026 GPT-5.6 Terra system card 100 Prompt injection, Connectors (4.2 table) Robustness
25 Jun 2026 GPT-5.6 Sol system card 100 Prompt injection, Connectors (4.2 table) Robustness
25 Jun 2026 GPT-5.6 Luna system card 99.9 Prompt injection, Connectors (4.2 table) Robustness
4 May 2026 GPT-5.5 Instant system card 99.2 Prompt injection attacks in connectors (Table 4) Robustness
23 Apr 2026 GPT-5.5 system card 96.3 Prompt injection attacks in connectors (Table 6, 4.2) Robustness
5 Mar 2026 GPT-5.4 system card 99.8 Prompt injection attacks in connectors (Table 4, 3.4) Robustness
11 Dec 2025 GPT-5.2 system card addendum 97.8 Prompt Injection, Agent JSK, gpt-5.2-thinking (Table 3) Robustness

More in Safety and alignment

All 444 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%
  • AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
  • MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy. Muse Spark Open 98.1%
  • 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. GPT-6 Astra 8.50%
  • Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%