Washington Post political lean evaluation
Models answer 30 US political questions in up to 30 words each. A reporter marks left- and right-leaning arguments and counts answers that present both.
The newspaper publishes 180 annotated answers across six models. A separate model judge scores five repeated runs per model.
- Made by
- The Washington Post
- Released
- June 2026
- 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
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-5.5 #21 on the capability index
- Last model added
- May 2026
- Scores refreshed
- November 20th, 2026
GPT-5.5 leads with 76.7% among models released by May 2026 GPT-5.5 leads with 76.7% among models released by May 2026
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in May 2026, 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 GPT-5.5 Reporter annotations 76.7%
- 2
DeepSeek-V4-Pro gpt-oss-20b judge, run 1 66.7%
- 3 arya gpt-oss-20b judge, run 5 46.7%
- 4 Claude Opus 4.8 Reporter annotations 43.3%
- 5 Grok 4.3 gpt-oss-20b judge, run 4 20%
- 6
Gemini 3.1 Pro gpt-oss-20b judge, run 3 10%
Source: Official Washington Post political lean evaluation leaderboard
Source: Official Washington Post political lean evaluation leaderboard
| # | Entry | Score | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | GPT-5.5 | 76.7% | Reporter annotations | Apr 2026 | Jun 2026 | Official leaderboard |
| 2 | | 66.7% | gpt-oss-20b judge, run 1 | Apr 2026 | Official leaderboard | |
| 3 | arya | 46.7% | gpt-oss-20b judge, run 5 | Official leaderboard | ||
| 4 | Claude Opus 4.8 | 43.3% | Reporter annotations | May 2026 | Jun 2026 | Official leaderboard |
| 5 | Grok 4.3 | 20% | gpt-oss-20b judge, run 4 | Apr 2026 | Official leaderboard | |
| 6 | | 10% | gpt-oss-20b judge, run 3 | Feb 2026 | Official leaderboard |
| # | Model | Score | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | GPT-5.5 | 76.7% | Reporter annotations | Apr 2026 | Jun 2026 | Official leaderboard |
| 2 | | 66.7% | gpt-oss-20b judge, run 1 | Apr 2026 | Official leaderboard | |
| 3 | Claude Opus 4.8 | 43.3% | Reporter annotations | May 2026 | Jun 2026 | Official leaderboard |
| 4 | Grok 4.3 | 20% | gpt-oss-20b judge, run 4 | Apr 2026 | Official leaderboard | |
| 5 | | 10% | gpt-oss-20b judge, run 3 | Feb 2026 | Official leaderboard |
Rankings by part
We read 3 parts from the benchmark’s own leaderboard: Left only, Right only and Both sides. Each ranks models on one part of the test. GPT-5.5 leads 1 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.
Left only
5 models we track have a score here.
- 1 GPT-5.5 80%
- 2
DeepSeek-V4-Pro 73.3%
- 3 Grok 4.3 ↑1 53.3%
- 4 Claude Opus 4.8 ↓1 43.3%
- 5
Gemini 3.1 Pro 10%
Right only
5 models we track have a score here.
- 1 Grok 4.3 ↑3 36.7%
- 2
DeepSeek-V4-Pro 20%
- 3 GPT-5.5 ↓2 10%
- 4 Claude Opus 4.8 ↓1 6.67%
- 5
Gemini 3.1 Pro 0%
Both sides
5 models we track have a score here.
- 1
Gemini 3.1 Pro ↑4 93.3%
- 2 Claude Opus 4.8 ↑1 60%
- 3 Grok 4.3 ↑1 36.7%
- 4
DeepSeek-V4-Pro ↓2 26.7%
- 5 GPT-5.5 ↓4 20%
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official Washington Post political lean evaluation leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| GPT-5.5 | Reporter annotations | 76.7% | |||
| GPT-5.5 | gpt-oss-20b judge, run 1 | 73.3% | |||
| GPT-5.5 | gpt-oss-20b judge, run 4 | 73.3% | |||
| GPT-5.5 | gpt-oss-20b judge, run 5 | 70% | |||
| GPT-5.5 | gpt-oss-20b judge, run 3 | 66.7% | |||
| gpt-oss-20b judge, run 1 | 66.7% | ||||
| GPT-5.5 | gpt-oss-20b judge, run 2 | 66.7% | |||
| Reporter annotations | 63.3% | ||||
| gpt-oss-20b judge, run 2 | 60% | ||||
| gpt-oss-20b judge, run 3 | 53.3% | ||||
| gpt-oss-20b judge, run 4 | 50% | ||||
| gpt-oss-20b judge, run 5 | 50% | ||||
| Claude Opus 4.8 | Reporter annotations | 43.3% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 1 | 43.3% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 2 | 33.3% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 5 | 33.3% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 4 | 30% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 3 | 26.7% | |||
| Grok 4.3 | gpt-oss-20b judge, run 4 | 20% | |||
| Grok 4.3 | gpt-oss-20b judge, run 3 | 16.7% | |||
| gpt-oss-20b judge, run 3 | 10% | ||||
| Grok 4.3 | gpt-oss-20b judge, run 2 | 10% | |||
| Grok 4.3 | gpt-oss-20b judge, run 5 | 10% | |||
| gpt-oss-20b judge, run 1 | 6.67% | ||||
| gpt-oss-20b judge, run 2 | 6.67% | ||||
| gpt-oss-20b judge, run 4 | 6.67% | ||||
| gpt-oss-20b judge, run 5 | 6.67% | ||||
| Reporter annotations | 6.67% | ||||
| Grok 4.3 | Reporter annotations | 6.67% | |||
| Grok 4.3 | gpt-oss-20b judge, run 1 | 3.33% |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official Washington Post political lean evaluation leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| GPT-5.5 | Reporter annotations | 76.7% | |||
| GPT-5.5 | gpt-oss-20b judge, run 1 | 73.3% | |||
| GPT-5.5 | gpt-oss-20b judge, run 4 | 73.3% | |||
| GPT-5.5 | gpt-oss-20b judge, run 5 | 70% | |||
| GPT-5.5 | gpt-oss-20b judge, run 3 | 66.7% | |||
| gpt-oss-20b judge, run 1 | 66.7% | ||||
| GPT-5.5 | gpt-oss-20b judge, run 2 | 66.7% | |||
| Reporter annotations | 63.3% | ||||
| gpt-oss-20b judge, run 2 | 60% | ||||
| gpt-oss-20b judge, run 3 | 53.3% | ||||
| gpt-oss-20b judge, run 4 | 50% | ||||
| gpt-oss-20b judge, run 5 | 50% | ||||
| Claude Opus 4.8 | Reporter annotations | 43.3% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 1 | 43.3% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 2 | 33.3% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 5 | 33.3% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 4 | 30% | |||
| Claude Opus 4.8 | gpt-oss-20b judge, run 3 | 26.7% | |||
| Grok 4.3 | gpt-oss-20b judge, run 4 | 20% | |||
| Grok 4.3 | gpt-oss-20b judge, run 3 | 16.7% | |||
| gpt-oss-20b judge, run 3 | 10% | ||||
| Grok 4.3 | gpt-oss-20b judge, run 2 | 10% | |||
| Grok 4.3 | gpt-oss-20b judge, run 5 | 10% | |||
| gpt-oss-20b judge, run 1 | 6.67% | ||||
| gpt-oss-20b judge, run 2 | 6.67% | ||||
| gpt-oss-20b judge, run 4 | 6.67% | ||||
| gpt-oss-20b judge, run 5 | 6.67% | ||||
| Reporter annotations | 6.67% | ||||
| Grok 4.3 | Reporter annotations | 6.67% | |||
| Grok 4.3 | gpt-oss-20b judge, run 1 | 3.33% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Washington Post political lean evaluation 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 Washington Post political lean evaluation 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
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 6 months ago, in May 2026, for Claude Opus 4.8. 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 official leaderboard (6 entries).
- Pass Models scored 5 models we track, plus 1 other entry on its leaderboard.
- Warning Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
- Pass Headroom The best is 76.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on washingtonpost.com: 30 rows for models we track and 1 other entry 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.