Score stability 1.1

Models: 21
Smallest CI width: 2.36
Largest CI width: 8.90
Most stable: GPT-6 Astra (high)

Two-dimensional view

Stability.

Each dot is one model. Lower means a better average question score. Further left means a narrower CI, so the model's results varied less across repeated rounds on these subjects.

Lower-left is betterNumbers match the table rank. Select a point to open the model's results.

1. GPT-6 Astra (high)13.67 questions · CI width 2.36
2. Gemini 4 Argon (high)13.67 questions · CI width 2.56
3. Claude Fable 5.1 (high)15.10 questions · CI width 2.69
4. Grok 4.7 (high)22.23 questions · CI width 2.87
5. Claude Opus 5 (high)18.07 questions · CI width 3.13
6. Grok 4.6 (high)17.13 questions · CI width 3.19
7. Gemini 3.8 Flash (high)14.87 questions · CI width 3.35
8. GPT-6 Sol (high)18.03 questions · CI width 3.48
9. Muse Spark 1.3 Contributor (high)18.37 questions · CI width 3.52
10. Claude Opus 5.5 (high)15.00 questions · CI width 3.53
11. Muse Wonder (high)19.63 questions · CI width 4.10
12. Gemini 3.7 Flash (high)16.57 questions · CI width 4.44
13. Claude Sonnet 5.5 (high)17.83 questions · CI width 4.69
14. GPT-6.1 Sol (high)15.23 questions · CI width 4.78
15. GPT-5.6 Luna (high)21.03 questions · CI width 5.05
16. gpt-oss-120B (high)22.23 questions · CI width 5.31
17. Claude Haiku 5.5 (high)20.00 questions · CI width 5.79
18. GPT-6 Luna (high)21.17 questions · CI width 5.97
19. GPT-5.6 Sol (high)18.07 questions · CI width 7.37
20. GLM-5.3-Flash (high)23.27 questions · CI width 7.38
21. MiniMax M3 (high)29.17 questions · CI width 8.90

GPT-6 Astra (high): score 13.67 questions; CI width 2.36 questions; stability rank 1. View results for GPT-6 Astra (high)
Gemini 4 Argon (high): score 13.67 questions; CI width 2.56 questions; stability rank 2. View results for Gemini 4 Argon (high)
Claude Fable 5.1 (high): score 15.10 questions; CI width 2.69 questions; stability rank 3. View results for Claude Fable 5.1 (high)
Grok 4.7 (high): score 22.23 questions; CI width 2.87 questions; stability rank 4. View results for Grok 4.7 (high)
Claude Opus 5 (high): score 18.07 questions; CI width 3.13 questions; stability rank 5. View results for Claude Opus 5 (high)
Grok 4.6 (high): score 17.13 questions; CI width 3.19 questions; stability rank 6. View results for Grok 4.6 (high)
Gemini 3.8 Flash (high): score 14.87 questions; CI width 3.35 questions; stability rank 7. View results for Gemini 3.8 Flash (high)
GPT-6 Sol (high): score 18.03 questions; CI width 3.48 questions; stability rank 8. View results for GPT-6 Sol (high)
Muse Spark 1.3 Contributor (high): score 18.37 questions; CI width 3.52 questions; stability rank 9. View results for Muse Spark 1.3 Contributor (high)
Claude Opus 5.5 (high): score 15.00 questions; CI width 3.53 questions; stability rank 10. View results for Claude Opus 5.5 (high)
Muse Wonder (high): score 19.63 questions; CI width 4.10 questions; stability rank 11. View results for Muse Wonder (high)
Gemini 3.7 Flash (high): score 16.57 questions; CI width 4.44 questions; stability rank 12. View results for Gemini 3.7 Flash (high)
Claude Sonnet 5.5 (high): score 17.83 questions; CI width 4.69 questions; stability rank 13. View results for Claude Sonnet 5.5 (high)
GPT-6.1 Sol (high): score 15.23 questions; CI width 4.78 questions; stability rank 14. View results for GPT-6.1 Sol (high)
GPT-5.6 Luna (high): score 21.03 questions; CI width 5.05 questions; stability rank 15. View results for GPT-5.6 Luna (high)
gpt-oss-120B (high): score 22.23 questions; CI width 5.31 questions; stability rank 16. View results for gpt-oss-120B (high)
Claude Haiku 5.5 (high): score 20.00 questions; CI width 5.79 questions; stability rank 17. View results for Claude Haiku 5.5 (high)
GPT-6 Luna (high): score 21.17 questions; CI width 5.97 questions; stability rank 18. View results for GPT-6 Luna (high)
GPT-5.6 Sol (high): score 18.07 questions; CI width 7.37 questions; stability rank 19. View results for GPT-5.6 Sol (high)
GLM-5.3-Flash (high): score 23.27 questions; CI width 7.38 questions; stability rank 20. View results for GLM-5.3-Flash (high)
MiniMax M3 (high): score 29.17 questions; CI width 8.90 questions; stability rank 21. View results for MiniMax M3 (high)

Stability ranking
Stability rank					95% CI
Rank 1	GPT-6 Astra (high)openai	2.36	1	13.67	12.49–14.85	100%
Rank 2	Gemini 4 Argon (high)google-vertex	2.56	2	13.67	12.39–14.95	100%
Rank 3	Claude Fable 5.1 (high)anthropic	2.69	5	15.10	13.76–16.44	93%
Rank 4	Grok 4.7 (high)xai	2.87	18	22.23	20.80–23.67	73%
Rank 5	Claude Opus 5 (high)anthropic	3.13	11	18.07	16.50–19.63	93%
Rank 6	Grok 4.6 (high)xai	3.19	8	17.13	15.54–18.73	100%
Rank 7	Gemini 3.8 Flash (high)google-ai-studio	3.35	3	14.87	13.19–16.54	100%
Rank 8	GPT-6 Sol (high)openai	3.48	10	18.03	16.29–19.77	100%
Rank 9	Muse Spark 1.3 Contributor (high)meta	3.52	13	18.37	16.61–20.13	97%
Rank 10	Claude Opus 5.5 (high)anthropic	3.53	4	15.00	13.24–16.76	97%
Rank 11	Muse Wonder (high)meta	4.10	14	19.63	17.58–21.68	97%
Rank 12	Gemini 3.7 Flash (high)google-ai-studio	4.44	7	16.57	14.35–18.79	93%
Rank 13	Claude Sonnet 5.5 (high)anthropic	4.69	9	17.83	15.49–20.18	90%
Rank 14	GPT-6.1 Sol (high)openai	4.78	6	15.23	12.84–17.62	97%
Rank 15	GPT-5.6 Luna (high)openai	5.05	16	21.03	18.51–23.56	90%
Rank 16	gpt-oss-120B (high)cerebras	5.31	19	22.23	19.58–24.89	67%
Rank 17	Claude Haiku 5.5 (high)anthropic	5.79	15	20.00	17.10–22.90	83%
Rank 18	GPT-6 Luna (high)openai	5.97	17	21.17	18.18–24.15	80%
Rank 19	GPT-5.6 Sol (high)openai	7.37	12	18.07	14.38–21.75	97%
Rank 20	GLM-5.3-Flash (high)z-ai	7.38	20	23.27	19.58–26.96	73%
Rank 21	MiniMax M3 (high)coreweave	8.90	21	29.17	24.71–33.62	50%

GPT-6 Astra (high)openai2.36CI width

Question score: 13.67
95% CI: 12.49–14.85
Solved: 100%

Gemini 4 Argon (high)google-vertex2.56CI width

Question score: 13.67
95% CI: 12.39–14.95
Solved: 100%

Claude Fable 5.1 (high)anthropic2.69CI width

Question score: 15.10
95% CI: 13.76–16.44
Solved: 93%

Grok 4.7 (high)xai2.87CI width

Question score: 22.23
95% CI: 20.80–23.67
Solved: 73%

Claude Opus 5 (high)anthropic3.13CI width

Question score: 18.07
95% CI: 16.50–19.63
Solved: 93%

Grok 4.6 (high)xai3.19CI width

Question score: 17.13
95% CI: 15.54–18.73
Solved: 100%

Gemini 3.8 Flash (high)google-ai-studio3.35CI width

Question score: 14.87
95% CI: 13.19–16.54
Solved: 100%

GPT-6 Sol (high)openai3.48CI width

Question score: 18.03
95% CI: 16.29–19.77
Solved: 100%

Muse Spark 1.3 Contributor (high)meta3.52CI width

Question score: 18.37
95% CI: 16.61–20.13
Solved: 97%

Claude Opus 5.5 (high)anthropic3.53CI width

Question score: 15.00
95% CI: 13.24–16.76
Solved: 97%

Muse Wonder (high)meta4.10CI width

Question score: 19.63
95% CI: 17.58–21.68
Solved: 97%

Gemini 3.7 Flash (high)google-ai-studio4.44CI width

Question score: 16.57
95% CI: 14.35–18.79
Solved: 93%

Claude Sonnet 5.5 (high)anthropic4.69CI width

Question score: 17.83
95% CI: 15.49–20.18
Solved: 90%

GPT-6.1 Sol (high)openai4.78CI width

Question score: 15.23
95% CI: 12.84–17.62
Solved: 97%

GPT-5.6 Luna (high)openai5.05CI width

Question score: 21.03
95% CI: 18.51–23.56
Solved: 90%

gpt-oss-120B (high)cerebras5.31CI width

Question score: 22.23
95% CI: 19.58–24.89
Solved: 67%

Claude Haiku 5.5 (high)anthropic5.79CI width

Question score: 20.00
95% CI: 17.10–22.90
Solved: 83%

GPT-6 Luna (high)openai5.97CI width

Question score: 21.17
95% CI: 18.18–24.15
Solved: 80%

GPT-5.6 Sol (high)openai7.37CI width

Question score: 18.07
95% CI: 14.38–21.75
Solved: 97%

GLM-5.3-Flash (high)z-ai7.38CI width

Question score: 23.27
95% CI: 19.58–26.96
Solved: 73%

MiniMax M3 (high)coreweave8.90CI width

Question score: 29.17
95% CI: 24.71–33.62
Solved: 50%

How it is calculated

CI width.

A smaller CI width means more consistent results across the repeated rounds.

FormulaCI width = upper 95% CI bound − lower 95% CI bound

Calculation detailsSteps, example, and limits

Calculate the model's 95% CI from the spread between repeated rounds of the same subject. Differences between subjects do not widen it.
Subtract the lower end from the upper end.
Sort the widths from smallest to largest.

Example, GPT-6 Astra (high): 14.85 − 12.49 = 2.36 questions.

Stable does not mean good. A model that scores poorly but consistently can rank well here. A model with a strong average and a wide CI ranks lower.

This is an approximate measure for these subjects only. It does not predict single rounds and is not a significance test between two models.