OvertonBench
Models answer questions about contested issues. Participants rate how well each answer represents their viewpoint. The score measures the fraction of opinion groups represented.
The authors rank eight models on 60 questions. The full leaderboard uses participant ratings. The paper also evaluates an automated judge.
- Made by
- Elinor Poole-Dayan and colleagues
- Released
- September 2025
- Kind
- Bias, fairness and politics
- Level
- Everyday
- Scored
- Graded by people
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Propensity
- Behavior
- Bias
- Subject
- Political science
- Language
- English
- Interaction
- Single turn
- Items from
- Written by crowdworkers
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- o4-mini #95 on the capability index
- Last model added
- May 2025
- Scores refreshed
- November 20th, 2026
DeepSeek-V3 leads with 0.42 among models released by May 2025
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in May 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
DeepSeek-V3 Mar 2025 0.42
- 2
DeepSeek-R1 May 2025 0.40
- 3
Llama 3.3 70B 0.40
- 4 GPT-4.1 0.40
- 5 o4-mini 0.39
- 6 Claude 3.7 Sonnet 0.39
- 7
Llama 4 Maverick 0.39
- 8
Gemma 3 27B 0.35
Source: Official OvertonBench leaderboard
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 0.42 0.4 to 0.5 | Mar 2025 | Official leaderboard | |
| 2 | | 0.40 0.4 to 0.5 | May 2025 | Official leaderboard | |
| 3 | | 0.40 ±0 | Dec 2024 | Official leaderboard | |
| 4 | GPT-4.1 | 0.40 ±0 | Apr 2025 | Official leaderboard | |
| 5 | o4-mini | 0.39 0.3 to 0.4 | Apr 2025 | Official leaderboard | |
| 6 | Claude 3.7 Sonnet | 0.39 ±0 | Feb 2025 | Official leaderboard | |
| 7 | | 0.39 0.3 to 0.4 | Apr 2025 | Official leaderboard | |
| 8 | | 0.35 ±0 | Mar 2025 | Official leaderboard |
Rankings by part
We read 2 parts from the benchmark’s own leaderboard: ModelSlant and PRISM. Each ranks models on one part of the test. o4-mini leads ModelSlant and DeepSeek-V3 leads PRISM. Arrows show how far a model moved from its overall rank among the models we track.
ModelSlant
8 models we track have a score here.
- 1 o4-mini ↑4 0.36
- 2
DeepSeek-R1 May 2025 0.31
- 3
Llama 3.3 70B 0.29
- 4
Gemma 3 27B ↑4 0.28
- 5 GPT-4.1 ↓1 0.27
- 6
Llama 4 Maverick ↑1 0.26
- 7 Claude 3.7 Sonnet ↓1 0.23
- 8
DeepSeek-V3 Mar 2025 ↓7 0.22
PRISM
8 models we track have a score here.
- 1
DeepSeek-V3 Mar 2025 0.49
- 2 Claude 3.7 Sonnet ↑4 0.45
- 3 GPT-4.1 ↑1 0.44
- 4
DeepSeek-R1 May 2025 ↓2 0.43
- 5
Llama 3.3 70B ↓2 0.43
- 6
Llama 4 Maverick ↑1 0.43
- 7 o4-mini ↓2 0.40
- 8
Gemma 3 27B 0.37
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official OvertonBench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official OvertonBench 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
- 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
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 18 months ago, in May 2025, for DeepSeek-R1. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (8 entries).
- Pass Models scored 8 models we track.
- Warning Newest score The newest came 18 months ago, in May 2025, for DeepSeek-R1.
- Note Frontier models None of the top 10 has a score. The strongest tested is o4-mini (#95).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on overtonbench.github.io: 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.