Benchmark Almanac

Prompt injection robustness

GPT-6 Astra 99.8% leads since Sep 2026 Only in system cards

Agents meet indirect prompt injection attacks inside tool outputs. Scored as defender success rate averaged per defender query.

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

Made by OpenAI Released Jul 2026 Jailbreaks and prompt injection Tasks kept private Last in a system card Oct 2026 Last model added Oct 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

GPT-6 Astra leads with 99.8%

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.

  1. 99.8 GPT-6 Astra
  2. 99.0 GPT-6 Sol
  3. 99.0 GPT-6.1 Sol
  4. 98.6 GPT-6 Luna
  5. 97.1 GPT-5.6 Luna
  6. 96.7 GPT-5.6 Terra
  7. 96.2 GPT-5.6 Sol
  1. 1 GPT-6 Astra 99.8%
  2. 2 GPT-6 Sol 99%
  3. 3 GPT-6.1 Sol 99%
  4. 4 GPT-6 Luna 98.6%
  5. 5 GPT-5.6 Luna 97.1%
  6. 6 GPT-5.6 Terra 96.7%
  7. 7 GPT-5.6 Sol 96.2%
#ModelScoreSettingDateSource
1 GPT-6 Astra 99.8% Sep 2026 Model card
2 GPT-6 Sol 99% Sep 2026 Model card
3 GPT-6.1 Sol 99% Sep 2026 Model card
4 GPT-6 Luna 98.6% Sep 2026 Model card
5 GPT-5.6 Luna 97.1% Jul 2026 Model card
6 GPT-5.6 Terra 96.7% Jul 2026 Model card
7 GPT-5.6 Sol 96.2% Jul 2026 Model card

2 models have held number one since its release in Jul 2026

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

OpenAI
Lab claims Passed 85% in Jul 2026, by GPT-5.6 Luna
GPT-6 Astra 99.8%

The best score passed 85% in Jul 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 Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6 Astra GPT-5.6 Luna 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6 Astra

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
29 Oct 2026 GPT-6 Luna system card addendum 94.5 GPT-RED indirect prompt-injection robustness, averaged across reasoning levels (Table 4) Robustness
26 Oct 2026 GPT-6.1 Terra system card addendum GPT-Red indirect prompt injection, defender success rate averaged per defender query, the plot is missing from the PDF and the text says Terra is similar to GPT-6 Astra and GPT-6.1 Sol (4.2) Robustness
7 Oct 2026 GPT-6 Sol system card addendum 97.1 GPT-RED indirect prompt-injection robustness, averaged across reasoning levels (Table 4) Robustness
7 Oct 2026 GPT-6 Luna system card addendum 95.8 GPT-RED indirect prompt-injection robustness, averaged across reasoning levels (Table 4) Robustness
29 Sep 2026 GPT-6.1 Sol system card addendum 99 Prompt injection robustness, printed chart label (Figure 3, 4.2) Robustness
22 Sep 2026 GPT-6 Sol system card 99 Prompt injection robustness, indirect prompt injection, printed chart label (Figure 59, A.3.2) Robustness
22 Sep 2026 GPT-6 Luna system card 98.6 Prompt injection robustness, indirect prompt injection, printed chart label (Figure 59, A.3.2) Robustness
3 Sep 2026 GPT-6 Astra system card 99.8 Indirect prompt injection robustness, defender success rate averaged per defender query, printed chart label (Figure 4, text 99.79%) Robustness
9 Jul 2026 GPT-5.6 Terra system card 96.7 GPT-Red Indirect Prompt Injections, attack success 3.32% per defender query, recorded as defender success (4.2, added August 3, 2026) Robustness
9 Jul 2026 GPT-5.6 Sol system card 96.2 GPT-Red Indirect Prompt Injections, attack success 3.77% per defender query, recorded as defender success (4.2, added August 3, 2026) Robustness
9 Jul 2026 GPT-5.6 Luna system card 97.1 GPT-Red Indirect Prompt Injections, attack success 2.94% per defender query, recorded as defender success (4.2, added August 3, 2026) 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%