Benchmark Almanac

Nuclear Decision-Making Benchmark

Gemini 3.1 Pro 81.3% leads since August 2026
Not in model cards

Models answer policy questions framed from different countries’ perspectives. The authors compare response distributions and agreement across repeated runs.

Eight models answered 151 scenarios. Higher five-run agreement measures consistency only. The questions have no single correct policy choice.

Released
August 2026
Kind
Bias, fairness and politics
Level
Expert
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Propensity
Behavior
Unreliability, Bias
Language
English
In and out
Text → Text
Interaction
Single turn
Items from
Written by experts
Refresh
One fixed set
Run by
Anyone
In views
Content safety
Strongest model tested
Gemini 3.1 Pro #40 on the capability index
Last model added
February 2026
Scores refreshed
November 20th, 2026
8
Models scored in its paper

Gemini 3.1 Pro leads with 81.3% among models released by February 2026

Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports no overall agreement percentage, so this is the plain mean of five-run agreement across its four domains. Where a model ran with several settings or harnesses, we keep its best.

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

  1. 1 Gemini 3.1 Pro 81.3%
  2. 2 Claude Sonnet 4.5 73.5%
  3. 3 GPT-5.2 69%
  4. 4 ERNIE-4.5-300B-A47B 65.8%
  5. 5 Qwen3-235B-A22B-Instruct Jul 2025 65.5%
  6. 6 GLM-4.7 65%
  7. 7 Llama 4 Maverick 61.3%
  8. 8 DeepSeek-V3.2 57.5%

Source: The Nuclear Decision-Making Benchmark paper

#ModelScoreSettingDateSource
1 Gemini 3.1 Pro 81.3% Feb 2026 Official leaderboard
2 Claude Sonnet 4.5 73.5% Sep 2025 Official leaderboard
3 GPT-5.2 69% Dec 2025 Official leaderboard
4 ERNIE-4.5-300B-A47B 65.8% Jun 2025 Official leaderboard
5 Qwen3-235B-A22B-Instruct Jul 2025 65.5% Jul 2025 Official leaderboard
6 GLM-4.7 65% Dec 2025 Official leaderboard
7 Llama 4 Maverick 61.3% Apr 2025 Official leaderboard
8 DeepSeek-V3.2 57.5% Dec 2025 Official leaderboard

Rankings by part

We read 4 parts from the benchmark’s paper, each ranking models on one part of the test, such as Escalation, Arms control and Non-proliferation. Gemini 3.1 Pro leads 2 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Escalation

8 models we track have a score here.

  1. 1 Gemini 3.1 Pro 72%
  2. 2 ERNIE-4.5-300B-A47B ↑2 49%
  3. 3 Claude Sonnet 4.5 ↓1 47%
  4. 4 GPT-5.2 ↓1 47%
  5. 5 Qwen3-235B-A22B-Instruct Jul 2025 37%
  6. 6 Llama 4 Maverick ↑1 36%
  7. 7 GLM-4.7 ↓1 31%
  8. 8 DeepSeek-V3.2 25%

Number one over time

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

Google DeepMind
Benchmark paper Not tested since Aug 2026

Source: The Nuclear Decision-Making Benchmark paper

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
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3.1 Pro Claude Sonnet 4.5 ERNIE-4.5-300B-A47B Llama 4 Maverick 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3.1 Pro Claude Sonnet 4.5

Source: Official Nuclear Decision-Making Benchmark 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 Newest score check. The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (8 entries).
  • Pass Models scored 8 models we track.
  • Warning Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3.1 Pro (#40).
  • Pass Headroom The best is 81.3%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 8 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