Benchmark Almanac

KORA

GPT-6 Astra 73.2% leads since Sep 2026 Not in model cards

Simulated child users hold 3-turn conversations probing 26 child-specific risks. Judge model rates replies failing, adequate or exemplary.

Public leaderboard is KORA V2 (repo tags 2.0.0 to 2.2.0, latest 2026-10-26). V3 changes scenario building and the judge and is in validation, so V2 and V3 scores will not be comparable. Relies on synthetic personas and an AI judge.

Made by KORA (non-profit) Released Feb 2026 Wellbeing and sycophancy All tasks public Last model added Oct 2026 Scores refreshed 2 Nov 2026
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.
korabench/benchmark 45 stars Leaderboard
42
Models scored on its own leaderboard

GPT-6 Astra leads with 73.2%

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.

42 models. Scroll sideways for the rest.

  1. 73.2 GPT-6 Astra
  2. 70.2 Claude Haiku 4.5
  3. 68.7 GPT-5.6 Sol
  4. 67.1 Claude Sonnet 4.6
  5. 67.0 GPT-5.5
  6. 65.8 Claude Opus 4.6
  7. 65.4 Claude Opus 4.8
  8. 63.6 Claude Sonnet 4.5
  9. 61.4 GPT-5.2
  10. 60.7 Mistral Large 4
  11. 60.0 Claude Opus 4.5
  12. 58.8 GPT-5.4
  13. 57.5 Claude Fable 5.1
  14. 57.2 Claude Opus 4.7
  15. 54.0 Claude 3.5 Haiku
  16. 53.8 Claude Fable 5
  17. 52.8 GPT-5.2 Instant
  18. 52.5 Kimi K3
  19. 51.5 GPT-5.3 Instant
  20. 48.4 GPT-5 nano
  21. 43.2 GPT-5 mini
  22. 40.6 GLM-5
  23. 38.5 Kimi K2.5
  24. 36.4 Gemini 2.5 Flash
  25. 31.6 Gemini 3.1 Pro
  26. 30.5 DeepSeek-V4-Pro
  27. 29.5 Gemini 3 Pro
  28. 29.1 Gemini 2.5 Pro
  29. 26.2 GPT-4.1
  30. 25.2 GLM-4.7
  31. 24.3 DeepSeek-V3.2
  32. 24.0 Gemini 3 Flash
  33. 23.6 GPT-4o
  34. 22.9 Llama 4 Maverick
  35. 22.6 gpt-oss-120b
  36. 18.0 Grok 3
  37. 17.7 Llama 4 Scout
  38. 16.1 gpt-oss-20b
  39. 14.8 Grok 4
  40. 14.3 Grok 4.1 Fast
  41. 10.8 Mistral Large 3
  42. 6.21 Ministral 3 14B
  1. 1 GPT-6 Astra high 73.2%
  2. 2 Claude Haiku 4.5 high 70.2%
  3. 3 GPT-5.6 Sol high 68.7%
  4. 4 Claude Sonnet 4.6 high 67.1%
  5. 5 GPT-5.5 high 67%
  6. 6 Claude Opus 4.6 high 65.8%
  7. 7 Claude Opus 4.8 high 65.4%
  8. 8 Claude Sonnet 4.5 high 63.6%
  9. 9 GPT-5.2 high 61.4%
  10. 10 Mistral Large 4 60.7%
  11. 11 Claude Opus 4.5 high 60%
  12. 12 GPT-5.4 high 58.8%
  13. 13 Claude Fable 5.1 high 57.5%
  14. 14 Claude Opus 4.7 high 57.2%
  15. 15 Claude 3.5 Haiku high 54%
  16. 16 Claude Fable 5 high 53.8%
  17. 17 GPT-5.2 Instant 52.8%
  18. 18 Kimi K3 52.5%
  19. 19 GPT-5.3 Instant 51.5%
  20. 20 GPT-5 nano high 48.4%
  21. 21 GPT-5 mini high 43.2%
  22. 22 GLM-5 40.6%
  23. 23 Kimi K2.5 38.5%
  24. 24 Gemini 2.5 Flash Sep 2025 36.4%
  25. 25 Gemini 3.1 Pro high 31.6%
  26. 26 DeepSeek-V4-Pro 30.5%
  27. 27 Gemini 3 Pro high 29.5%
  28. 28 Gemini 2.5 Pro Jun 2025 29.1%
  29. 29 GPT-4.1 26.2%
  30. 30 GLM-4.7 25.2%
  31. 31 DeepSeek-V3.2 24.3%
  32. 32 Gemini 3 Flash high 24%
  33. 33 GPT-4o Mar 2025 23.6%
  34. 34 Llama 4 Maverick 22.9%
  35. 35 gpt-oss-120b 22.6%
  36. 36 Grok 3 18%
  37. 37 Llama 4 Scout 17.7%
  38. 38 gpt-oss-20b 16.1%
  39. 39 Grok 4 14.8%
  40. 40 Grok 4.1 Fast 14.3%
  41. 41 Mistral Large 3 10.8%
  42. 42 Ministral 3 14B 6.21%
#ModelScoreSettingDateSource
1 GPT-6 Astra 73.2% high Sep 2026 Official leaderboard
2 Claude Haiku 4.5 70.2% high Oct 2025 Official leaderboard
3 GPT-5.6 Sol 68.7% high Jul 2026 Official leaderboard
4 Claude Sonnet 4.6 67.1% high Feb 2026 Official leaderboard
5 GPT-5.5 67% high Apr 2026 Official leaderboard
6 Claude Opus 4.6 65.8% high Feb 2026 Official leaderboard
7 Claude Opus 4.8 65.4% high May 2026 Official leaderboard
8 Claude Sonnet 4.5 63.6% high Sep 2025 Official leaderboard
9 GPT-5.2 61.4% high Dec 2025 Official leaderboard
10 Mistral Large 4 60.7% Oct 2026 Official leaderboard
11 Claude Opus 4.5 60% high Nov 2025 Official leaderboard
12 GPT-5.4 58.8% high Mar 2026 Official leaderboard
13 Claude Fable 5.1 57.5% high Sep 2026 Official leaderboard
14 Claude Opus 4.7 57.2% high Apr 2026 Official leaderboard
15 Claude 3.5 Haiku 54% high Oct 2024 Official leaderboard
16 Claude Fable 5 53.8% high Jun 2026 Official leaderboard
17 GPT-5.2 Instant 52.8% Dec 2025 Official leaderboard
18 Kimi K3 52.5% Jul 2026 Official leaderboard
19 GPT-5.3 Instant 51.5% Mar 2026 Official leaderboard
20 GPT-5 nano 48.4% high Aug 2025 Official leaderboard
21 GPT-5 mini 43.2% high Aug 2025 Official leaderboard
22 GLM-5 40.6% Feb 2026 Official leaderboard
23 Kimi K2.5 38.5% Jan 2026 Official leaderboard
24 Gemini 2.5 Flash Sep 2025 36.4% Sep 2025 Official leaderboard
25 Gemini 3.1 Pro 31.6% high Feb 2026 Official leaderboard
26 DeepSeek-V4-Pro 30.5% Apr 2026 Official leaderboard
27 Gemini 3 Pro 29.5% high Nov 2025 Official leaderboard
28 Gemini 2.5 Pro Jun 2025 29.1% Jun 2025 Official leaderboard
29 GPT-4.1 26.2% Apr 2025 Official leaderboard
30 GLM-4.7 25.2% Dec 2025 Official leaderboard
31 DeepSeek-V3.2 24.3% Dec 2025 Official leaderboard
32 Gemini 3 Flash 24% high Dec 2025 Official leaderboard
33 GPT-4o Mar 2025 23.6% Mar 2025 Official leaderboard
34 Llama 4 Maverick 22.9% Apr 2025 Official leaderboard
35 gpt-oss-120b 22.6% Aug 2025 Official leaderboard
36 Grok 3 18% Apr 2025 Official leaderboard
37 Llama 4 Scout 17.7% Apr 2025 Official leaderboard
38 gpt-oss-20b 16.1% Aug 2025 Official leaderboard
39 Grok 4 14.8% Jul 2025 Official leaderboard
40 Grok 4.1 Fast 14.3% Nov 2025 Official leaderboard
41 Mistral Large 3 10.8% Dec 2025 Official leaderboard
42 Ministral 3 14B 6.21% Dec 2025 Official leaderboard

2 models have held number one since its release in Feb 2026

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

Anthropic OpenAI
Own leaderboard 2 leaders
GPT-6 Astra 73.2%

Scores over time

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

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-6 Astra Claude Haiku 4.5 Claude Sonnet 4.5 Claude 3.5 Haiku 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-6 Astra Claude Sonnet 4.5

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

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%