WeirdML

Unusual machine-learning tasks that require writing working training code.

Published by

WeirdMLdata via Epoch AI Benchmarking Hub

Category

Coding

Index weight

0.5

Models

150

Data as of

9 Sept 2026

CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.

1. 1Claude Fable 5.192.9%
2. 2GPT-6 Astra92.9%
3. 3Claude Fable 5.192.3%
4. 4Claude Fable 591.9%
5. 5Claude Opus 591.8%
6. 6Claude Opus 591.6%
7. 7GPT-5.6 Sol89.4%
8. 8GPT-5.6 Sol88.8%
9. 9Claude Fable 587.8%
10. 10GPT-5.6 Sol87.0%
11. 11Claude Opus 586.3%
12. 12GPT-5.584.9%
13. 13GPT-5.583.9%
14. 14Claude Opus 4.882.9%
15. 15Kimi K382.6%

SearchAll providersAlibabaAnthropicDeepSeekGoogleInceptionMetaMiniMaxMistral AIMoonshot AINVIDIAOpenAIxAIZhipu AIOpen weights150 of 150

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Claude Fable 5.1Anthropic	92.9%	–	70.2	–
2	GPT-6 AstraOpenAI	92.9%	–	69.5	–
3	Claude Fable 5.1Anthropic	92.3%	–	69.7	–
4	Claude Fable 5Anthropic	91.9%	–	65.8	–
5	Claude Opus 5Anthropic	91.8%	–	66.8	–
6	Claude Opus 5Anthropic	91.6%	–	69.3	–
7	GPT-5.6 SolOpenAI	89.4%	–	–	–
8	GPT-5.6 SolOpenAI	88.8%	–	68.0	–
9	Claude Fable 5Anthropic	87.8%	–	63.9	–
10	GPT-5.6 SolOpenAI	87.0%	–	65.5	–
11	Claude Opus 5Anthropic	86.3%	–	70.0	–
12	GPT-5.5OpenAI	84.9%	–	63.7	–
13	GPT-5.5OpenAI	83.9%	–	66.3	–
14	Claude Opus 4.8Anthropic	82.9%	–	–	–
15	Kimi K3Moonshot AIopen	82.6%	–	62.7	–
16	GPT-5.3 CodexOpenAI	79.3%	–	62.4	–
17	GPT-5.6 TerraOpenAI	78.3%	–	63.1	–
18	Claude Opus 4.6Anthropic	78.0%	–	61.3	–
19	Claude Opus 4.6Anthropic	77.9%	–	63.1	–
20	GPT-5.3 CodexOpenAI	77.9%	–	–	–
21	GPT-5.4OpenAI	77.7%	–	61.3	–
22	Claude Opus 4.7Anthropic	76.4%	–	61.3	–
23	Claude Opus 4.7Anthropic	76.4%	–	65.5	–
24	Claude Opus 4.8Anthropic	76.0%	–	–	–
25	Claude Opus 4.7Anthropic	75.5%	–	59.3	–
26	GLM-5.3Zhipu AIopen	75.4%	–	59.9	–
27	GPT-5.2OpenAI	72.2%	–	57.3	–
28	Gemini 3.1 ProGoogle	72.1%	–	63.9	–
29	Claude Opus 4.8Anthropic	70.5%	–	–	–
30	GLM-5.2Zhipu AIopen	70.1%	–	57.6	–
31	Gemini 3 ProGoogle	69.9%	–	61.6	–
32	Claude Sonnet 5Anthropic	68.8%	–	60.5	–
33	GLM-5.2Zhipu AIopen	67.3%	–	–	–
34	Grok 4.6xAI	67.3%	–	58.7	–
35	GPT-5.5OpenAI	67.2%	–	56.2	–
36	DeepSeek V4 ProDeepSeekopen	66.2%	max, 0813	56.1	–
37	Claude Sonnet 4.6Anthropic	66.1%	–	51.8	–
38	Claude Opus 4.5Anthropic	63.7%	–	58.0	–
39	GPT-5.2OpenAI	63.4%	–	58.5	–
40	DeepSeek V4 FlashDeepSeekopen	63.0%	max, 0731	54.7	–
41	Gemini 3.5 FlashGoogle	62.6%	–	60.4	–
42	Gemini 3 FlashGoogle	61.6%	–	56.6	–
43	GPT-5.6 LunaOpenAI	60.9%	–	58.1	–
44	GPT-5.1OpenAI	60.8%	–	53.8	–
45	GPT-5OpenAI	60.7%	–	54.5	–
46	GPT-5 ProOpenAI	60.4%	–	–	–
47	Muse Spark 1.2Meta	60.3%	–	61.3	–
48	GPT-5.4 miniOpenAI	60.3%	–	52.9	–
49	o3-proOpenAI	58.2%	–	54.4	–
50	GPT-5.4OpenAI	57.4%	–	53.7	–
51	GPT-5.4 ProOpenAI	57.4%	–	–	–
52	GLM-5.1Zhipu AIopen	57.1%	glm-5.1	60.1	–
53	DeepSeek V4 FlashDeepSeekopen	57.0%	high, 0731	59.3	–
54	Gemini 3.6 FlashGoogle	56.1%	–	58.6	–
55	Kimi K2.6Moonshot AIopen	55.9%	kimi-k2.6	60.6	–
56	GPT-5-CodexOpenAI	54.5%	–	58.1	–
57	GPT-5-CodexOpenAI	54.5%	–	–	–
58	Kimi K2.7 CodeMoonshot AIopen	54.1%	–	56.2	–
59	Gemini 2.5 ProGoogle	54.0%	–	54.1	–
60	GPT-5 miniOpenAI	52.7%	–	52.3	–

Show 90 more