Benchmark Almanac

FrontierBreach

Gemini 4 Carbon 2.40% leads since November 2026
Not in model cards

Agents perform simulated enterprise tasks while processing untrusted material. The benchmark checks whether each agent resists instructions that conflict with its assigned task.

The leaderboard reports attack success rates with confidence intervals. Lower rates indicate greater resistance.

Made by
PromptArmor
Released
November 2026
Kind
Jailbreaks and prompt injection
Level
Expert
Scored
Checked automatically
Answers
Agent task
Measures
Propensity
Language
English
In and out
Text → Actions, Text
Interaction
Agentic
Environment
A simulator
Refresh
One fixed set
Run by
A host
In views
Agents, Robotics
Strongest model tested
Claude Fable 5.2 #2 on the capability index
Last model added
November 2026
Scores refreshed
November 20th, 2026
22
Models scored on PromptArmor’s leaderboard

Gemini 4 Carbon leads with 2.40%

Best score per model from PromptArmor’s leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best.

  1. 1 Gemini 4 Carbon 2.40%
  2. 2 Gemini 4 Argon 4.30%
  3. 3 GPT-6 Astra 7.60%
  4. 4 Claude Opus 5 10%
  5. 5 Claude Fable 5.2 10%
  6. 6 Muse Wonder 15.2%
  7. 7 Gemini 3.6 Flash 22.4%
  8. 8 Inkling 2 Small 30%
  9. 9 Qwen3.8 Max (0902) 30.5%
  10. 10 Kimi K3 31.4%
  11. 11 GPT-5.6 Sol 33.3%
  12. 12 Grok 4.7 33.3%
  13. 13 Qwen3.7-Max 36.2%
  14. 14 GLM-5.3 37.1%
  15. 15 Muse Spark 1.1 38.6%
  16. 16 GLM-5.2 39%
  17. 17 Inkling 39%
  18. 18 DeepSeek V4 Flash 0731 49.5%
  19. 19 Kimi K3.1 Pro 50%
  20. 20 DeepSeek V4 Pro 0813 51.4%
  21. 21 Beam 1.0 72.9%
  22. 22 Grok 4.5 91.4%

Source: PromptArmor’s leaderboard

#ModelScore ±SettingDateSource
1 Gemini 4 Carbon 2.40% ±2.2 Oct 2026 Official leaderboard
2 Gemini 4 Argon 4.30% ±2.9 Sep 2026 Official leaderboard
3 GPT-6 Astra 7.60% ±3.8 Sep 2026 Official leaderboard
4 Claude Opus 5 10% ±4.2 Jul 2026 Official leaderboard
5 Claude Fable 5.2 10% ±4.2 Nov 2026 Official leaderboard
6 Muse Wonder 15.2% ±5.1 Oct 2026 Official leaderboard
7 Gemini 3.6 Flash 22.4% ±5.9 Jul 2026 Official leaderboard
8 Inkling 2 Small 30% ±6.5 Nov 2026 Official leaderboard
9 Qwen3.8 Max (0902) 30.5% ±6.5 Sep 2026 Official leaderboard
10 Kimi K3 31.4% ±6.6 Jul 2026 Official leaderboard
11 GPT-5.6 Sol 33.3% ±6.7 Jul 2026 Official leaderboard
12 Grok 4.7 33.3% ±6.7 Sep 2026 Official leaderboard
13 Qwen3.7-Max 36.2% ±6.8 May 2026 Official leaderboard
14 GLM-5.3 37.1% ±6.8 Aug 2026 Official leaderboard
15 Muse Spark 1.1 38.6% ±6.9 Jul 2026 Official leaderboard
16 GLM-5.2 39% ±6.9 Jun 2026 Official leaderboard
17 Inkling 39% ±6.9 Jul 2026 Official leaderboard
18 DeepSeek V4 Flash 0731 49.5% ±7.1 Jul 2026 Official leaderboard
19 Kimi K3.1 Pro 50% ±7.1 Oct 2026 Official leaderboard
20 DeepSeek V4 Pro 0813 51.4% ±7.1 Aug 2026 Official leaderboard
21 Beam 1.0 72.9% ±6.3 Oct 2026 Official leaderboard
22 Grok 4.5 91.4% ±4 Jul 2026 Official leaderboard

Number one over time

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

Google DeepMind
PromptArmor leaderboard 1 leader

Source: PromptArmor’s leaderboard

Scores over time

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

Plot with other benchmarks
On PromptArmor’s leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Gemini 4 Carbon GPT-6 Astra Gemini 3.6 Flash Kimi K3 Qwen3.7-Max 0% 20% 40% 60% 80% 100% 2026 2027 Gemini 4 Carbon Claude Opus 5

Source: PromptArmor’s leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Safety and alignment

All 636 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%
  • Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior. Inkling 2 Mini 8.30%
  • Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions. Inkling 2 Small 4.70%
  • Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior. Inkling 2 Mini 9.80%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark

Catalog record 1 warning 1 pass
  • Warning Task access We don’t know whether the tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (22 entries).
  • Pass Models scored 22 models we track.
  • Pass Newest score The newest is from November 2026, for Claude Fable 5.2.
  • Pass Frontier models 4 of the top 10 have a score, led by Claude Fable 5.2 (#2).
  • Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
  • Pass Importer We read the board on PromptArmor: 22 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check