Benchmark Almanac

Anthropic political even-handedness eval

Claude Opus 5.5 99.4% leads since Sep 2026 Only in system cards

Models answer paired prompts that take opposing sides on 150 US political topics. A grader scores even-handedness, refusals and acknowledgment of opposing views.

Made by Anthropic Released Nov 2025 Bias, fairness and politics All tasks public Last in a system card Oct 2026 Last model added Oct 2026
Level
Everyday Most adults, without training
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
anthropics/political-neutrality-eval 142 stars anthropic.com
13
Models scored in lab cards
15
System cards report it, from 1 lab

Claude Opus 5.5 leads with 99.4%

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.

13 models. Scroll sideways for the rest.

  1. 99.4 Claude Opus 5.5
  2. 98.4 Claude Sonnet 4.6
  3. 98.3 Claude Fable 5.1
  4. 98.2 Claude Opus 4.6
  5. 97.9 Claude Sonnet 5.5
  6. 97.7 Claude Haiku 5.5
  7. 97.6 Claude Fable 5
  8. 96.7 Claude Opus 4.8
  9. 96.4 Claude Sonnet 5
  10. 96.3 Claude Opus 5
  11. 96.0 Claude Opus 4.5
  12. 95.1 Claude Opus 4.7
  13. 94.5 Claude Mythos Preview
  1. 1 Claude Opus 5.5 99.4%
  2. 2 Claude Sonnet 4.6 98.4%
  3. 3 Claude Fable 5.1 98.3%
  4. 4 Claude Opus 4.6 98.2%
  5. 5 Claude Sonnet 5.5 97.9%
  6. 6 Claude Haiku 5.5 97.7%
  7. 7 Claude Fable 5 97.6%
  8. 8 Claude Opus 4.8 96.7%
  9. 9 Claude Sonnet 5 96.4%
  10. 10 Claude Opus 5 96.3%
  11. 11 Claude Opus 4.5 96%
  12. 12 Claude Opus 4.7 95.1%
  13. 13 Claude Mythos Preview 94.5%
#ModelScoreSettingDateSource
1 Claude Opus 5.5 99.4% Sep 2026 Model card
2 Claude Sonnet 4.6 98.4% Feb 2026 Model card
3 Claude Fable 5.1 98.3% Sep 2026 Model card
4 Claude Opus 4.6 98.2% Feb 2026 Model card
5 Claude Sonnet 5.5 97.9% Sep 2026 Model card
6 Claude Haiku 5.5 97.7% Oct 2026 Model card
7 Claude Fable 5 97.6% Jun 2026 Model card
8 Claude Opus 4.8 96.7% May 2026 Model card
9 Claude Sonnet 5 96.4% Jun 2026 Model card
10 Claude Opus 5 96.3% Jul 2026 Model card
11 Claude Opus 4.5 96% Nov 2025 Model card
12 Claude Opus 4.7 95.1% Apr 2026 Model card
13 Claude Mythos Preview 94.5% Apr 2026 Model card

4 models have held number one since its release in Nov 2025

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

Anthropic
Lab claims Passed 85% in Nov 2025, by Claude Opus 4.5
Claude Opus 5.5 99.4%

The best score passed 85% in Nov 2025

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% 2026 Near the ceiling (85%) Claude Opus 5.5 Claude Sonnet 4.6 Claude Opus 4.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Claude Opus 4.5

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.

15 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
7 Oct 2026 Claude Haiku 5.5 system card 97.7 Political even-handedness, API without system prompt (claude.ai 99.1). Opposing perspectives 35.9, refusals 3.3 on API (4.4.1) Bias
28 Sep 2026 Claude Sonnet 5.5 system card 97.9 Political even-handedness, API without system prompt (claude.ai 99.0%), opposing perspectives 41.4% (4.4.1, Figure 4.4.1.A) Bias
22 Sep 2026 Claude Opus 5.5 system card 99.4 Political even-handedness, 1,350 prompt pairs, API without a system prompt (claude.ai 99.5%), opposing perspectives 26.9% on the API (4.4.1, Figures 4.4.1.A to C) Bias
1 Sep 2026 Claude Fable 5.1 system card 98.3 Pairwise political bias, even-handedness, API without a system prompt (claude.ai 98.8%); opposing perspectives 25.9% (claude.ai 80.5%), refusals 2.1% (claude.ai 5.0%) (Figures 4.4.1.A to 4.4.1.C) Bias
24 Jul 2026 Claude Opus 5 system card 96.3 Political even-handedness, 1,350 prompt pairs, API without a system prompt (claude.ai 98.9%). Opposing perspectives 46.9% and refusals 3.0% on the API (75.8% and 4.2% on claude.ai), printed on charts (4.4.1, Figures 4.4.1.A and B) Bias
30 Jun 2026 Claude Sonnet 5 system card 96.4 Political even-handedness, 1,350 prompt pairs, public system prompt applied, opposing perspectives 54.6%, refusals 4.7% (4.4.1, Figure 4.4.1.A) Bias
9 Jun 2026 Claude Fable 5 system card 97.6 Pairwise political bias, even-handedness, public system prompt applied; opposing perspectives 70.0%, refusals 13.3%, printed on chart (Figure 4.4.1.A) Bias
28 May 2026 Claude Opus 4.8 system card 96.7 Political even-handedness, 1,350 prompt pairs, public system prompt applied, thinking on and off pooled. Opposing perspectives 66.4%, refusals 7.2%, printed on chart (4.4.1, Figure 4.4.1.A) Bias
16 Apr 2026 Claude Opus 4.7 system card 95.1 Pairwise political bias, even-handedness with the public system prompt, no thinking, printed on chart. Opposing perspectives 40.1, refusals 6.7 (4.5.1, Figure 4.5.1.A) Bias
7 Apr 2026 Claude Mythos Preview system card 94.5 Political even-handedness with public system prompt, no thinking; opposing perspectives 47.0, refusals 13.5 (Table 8.2.1.A) Bias
17 Feb 2026 Claude Sonnet 4.6 system card 98.4 Political even-handedness with system prompt, no thinking; opposing perspectives 32.1, refusals 4.5 (Table 3.5.1.A) Bias
5 Feb 2026 Claude Opus 4.6 system card 98.2 Political even-handedness, 1,350 prompt pairs, public system prompt, no thinking; opposing perspectives 44.6, refusals 4.5 (Table 3.5.1.A) Bias
24 Nov 2025 Claude Opus 4.5 system card 96 Political even-handedness, 1,350 prompt pairs, public system prompt, no extended thinking (Figure 3.5.1.A 0.96; opposing perspectives 0.40, refusals 0.04) Bias
15 Oct 2025 Claude Haiku 4.5 system card Political bias, paired prompts, substantial asymmetry rate 5.3% in standard thinking and 10% across standard and extended thinking, lower is better, no even-handedness score (2.5.1) Bias
29 Sep 2025 Claude Sonnet 4.5 system card Political bias, paired prompts, substantial asymmetry rate 3.3% averaged over thinking on and off (1.3% with extended thinking, 5.3% without), lower is better, no even-handedness score (2.5.1) Bias

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%