Rangliste

Research Rankings

Wissenschaftliche Literatur und Expertenwissen: HLE, GPQA Diamond, MMLU-Pro.

Gesamt700 Coding399 Mathematik53 Reasoning413 Agentic55 Wissen412 Multimodal382 Coding-Agenten55 Wissenschaft82 Mehrsprachig12 Schreiben61 Chat & Assistenten61 Research85 Tool-Nutzung32 Langkontext82 Tempo402 Preis-Leistung401 Open Weights170

85 Modelle

#	Modell	Podium Score	GPQA Diamond	Humanity’s Last Exam	MMLU-Pro	Geschwindigkeit
01	Seed 2.0 Pro ByteDance	88.9	88.9%	—	—	40 t/s
02	GPT-5.1 OpenAI	88.1	88.1%	—	—	60 t/s
03	Claude Opus 4.5 Anthropic	87.0	87%	—	—	50 t/s
4	ERNIE 5.1 Baidu	87.0	—	—	87%	60 t/s
5	Qwen3.5 122B-A10B Alibaba · Offen	86.7	86.6%	—	86.7%	126 t/s
6	GPT-5 OpenAI	85.7	85.7%	—	—	60 t/s
7	Gemma 4 31B Google · Offen	85.2	—	—	85.2%	35 t/s
8	DeepSeek V3.2 DeepSeek · Offen	85.0	—	—	85%	60 t/s
9	Qwen3.8 Flash Next Alibaba · Offen	81.0	81%	—	—	168 t/s
10	Claude Mythos Preview Anthropic · reasoning · preview	79.7	94.6%	64.7%	—	80 t/s
11	GLM-5.3 Flash Z.ai · Offen	78.5	78.5%	—	—	182 t/s
12	Claude Opus 4.8 Anthropic	75.8	93.6%	57.9%	—	50 t/s
13	Gemini 3 Pro Google	75.8	91.9%	45.8%	89.8%	120 t/s
14	DeepSeek V4 Pro DeepSeek · Offen	75.3	90.1%	48.2%	87.5%	66 t/s
15	Kimi K3 Moonshot AI · Offen · reasoning	74.8	93.5%	56%	—	37 t/s
16	GPT-6 Astra OpenAI · reasoning	74.6	96%	53.2%	—	82 t/s
17	Claude Opus 4.7 Anthropic	74.5	94.2%	54.7%	—	80 t/s
18	Qwen3.7 Max Alibaba	74.5	92.4%	41.4%	89.6%	203 t/s
19	Muse Spark Meta	74.0	89.5%	58.4%	—	80 t/s
20	DeepSeek V4 Flash DeepSeek · Offen	73.1	88.1%	45.1%	86.2%	113 t/s
21	GLM-5.2 Z.ai · Offen	73.0	91.2%	54.7%	—	164 t/s
22	Claude Fable 5 Anthropic · reasoning	72.9	92.6%	53.3%	—	71 t/s
23	Claude Opus 5 Anthropic	72.9	93.2%	52.6%	—	60 t/s
24	Gemini 3.1 Pro Google	72.9	94.3%	51.4%	—	136 t/s
25	GPT-5.5 OpenAI	72.9	93.6%	52.2%	—	60 t/s
26	Claude Opus 4.6 Anthropic	72.2	91.3%	53.1%	—	50 t/s
27	Qwen3.7 Plus Alibaba	71.2	90.3%	34.7%	88.5%	54 t/s
28	GPT-5.6 Sol OpenAI	70.9	94.6%	47.2%	—	72 t/s
29	Grok 4 Heavy xAI · preview	69.6	88.4%	50.7%	—	50 t/s
30	Claude Sonnet 4.6 Anthropic	69.5	89.9%	49%	—	50 t/s
31	Qwen3.6 Plus Alibaba	69.2	90.4%	28.8%	88.5%	55 t/s
32	Qwen3.5 397B-A17B Alibaba · Offen	68.3	88.4%	28.7%	87.8%	71 t/s
33	Qwen3.8 Max Alibaba	68.1	92.6%	43.6%	—	55 t/s
34	Granite 4.2 30B Instruct IBM · Offen	68.0	68%	—	—	78 t/s
35	Muse Spark 1.1 Meta	67.5	89.8%	45.1%	—	208 t/s
36	GPT-5.6 Terra OpenAI	67.4	92.9%	41.8%	—	127 t/s
37	GLM-5.1 Z.ai · Offen	67.0	67%	—	—	74 t/s
38	Gemini 3 Flash Google	67.0	90.4%	43.5%	—	120 t/s
39	Claude Sonnet 4.6 (max) Anthropic	67.0	67%	—	—	50 t/s
40	Nemotron 3 Ultra 550B NVIDIA · Offen	66.8	87%	26.6%	86.8%	142 t/s
41	Grok 4.5 xAI	66.7	93%	40.3%	—	59 t/s
42	Gemini 3.5 Flash Google	66.6	92.2%	41%	—	267 t/s
43	GPT-5.4 OpenAI	66.3	92.8%	39.8%	—	60 t/s
44	GPT-5.3 Codex OpenAI	65.7	91.5%	39.9%	—	118 t/s
45	Gemini 3.6 Flash Google	65.6	92.8%	38.3%	—	233 t/s
46	MiniMax M3 MiniMax	65.0	92.9%	37.1%	—	93 t/s
47	GPT-5.6 Luna OpenAI	64.8	92.3%	37.2%	—	175 t/s
48	MiniMax-M2.7 MiniMax	64.0	64%	—	—	40 t/s
49	GPT-5.2 OpenAI	63.5	92.4%	34.5%	—	90 t/s
50	Kimi K2.6 Moonshot AI · Offen	63.5	90.5%	36.4%	—	40 t/s
51	DeepSeek V4.1 Flash DeepSeek · Offen · reasoning	62.0	62%	—	—	214 t/s
52	Kimi K2.7 Code Moonshot AI · Offen	61.2	89.6%	32.8%	—	39 t/s
53	Hunyuan Hy3 Tencent	61.0	90.4%	31.6%	—	69 t/s
54	MiMo V2.5 Pro Xiaomi · Offen	60.2	86.6%	33.8%	—	65 t/s
55	Inkling Thinking Machines	58.5	87.2%	29.7%	—	81 t/s
56	Granite 4.2 8B Instruct IBM · Offen	54.0	54%	—	—	115 t/s
57	Mistral Medium 3.5 Mistral	51.0	51%	—	—	60 t/s
58	Agnes 3.0 Flash Agnes · reasoning	49.0	49%	—	—	235 t/s
59	Claude Sonnet 5 Anthropic	48.9	80%	17.8%	—	96 t/s
60	Claude 4.5 Haiku Anthropic	48.0	48%	—	—	80 t/s
61	Nova 2.0 Pro Preview Amazon · preview	46.0	46%	—	—	40 t/s
62	Agnes 2.5 Pro Beta Agnes	45.0	45%	—	—	65 t/s
63	GPT-oss-120B OpenAI · Offen	43.0	43%	—	—	240 t/s
64	Granite 4.2 3B Instruct IBM · Offen	42.0	42%	—	—	148 t/s
65	DeepSeek V4 Flash Vision DeepSeek	36.0	36%	—	—	120 t/s
66	G9v3-39A5B AI9Stars · Offen	36.0	36%	—	—	60 t/s
67	G9v3-3B AI9Stars · Offen	36.0	36%	—	—	60 t/s
68	GLM-5.3 Zhipu AI	36.0	36%	—	—	90 t/s
69	Grok 4.6 (low) xAI	36.0	36%	—	—	59 t/s
70	Grok 4.6 (medium) xAI	36.0	36%	—	—	59 t/s
71	Grok 4.6 (xhigh) xAI	36.0	36%	—	—	62 t/s
72	HyperNova 60B Multiverse · Offen	36.0	36%	—	—	367 t/s
73	KAT-Coder-Pro V2 Kwai	36.0	36%	—	—	103 t/s
74	LFM2.5-2.6B Liquid AI · Offen	36.0	36%	—	—	60 t/s
75	LFM2.5-8B-A1B Liquid AI · Offen	36.0	36%	—	—	338 t/s
76	LFM2.5-VL-1.6B Liquid AI · Offen	36.0	36%	—	—	395 t/s
77	LongCat 2.0 LongCat · Offen	36.0	36%	—	—	42 t/s
78	Magistral Small 1.2 Mistral · Offen	36.0	36%	—	—	60 t/s
79	Qwen3.8 27B (xhigh) Alibaba · Offen	36.0	36%	—	—	53 t/s
80	Qwen3.8 27B (low) Alibaba · Offen	36.0	36%	—	—	64 t/s
81	Qwen3.8 27B (medium) Alibaba · Offen	36.0	36%	—	—	62 t/s
82	Qwen3.8 27B (Non-reasoning) Alibaba · Offen · reasoning	36.0	36%	—	—	62 t/s
83	Solar Pro 3 Upstage	34.0	34%	—	—	40 t/s
84	GPT-oss-20B OpenAI · Offen	32.0	32%	—	—	90 t/s
85	K2 Think V2 K2	31.0	31%	—	—	40 t/s

Der Podium Score ist eine gewichtete Mischung aus fünf Ranglisten. Siehe Methodik für Details.