Ranking

Ranking: Agentes

Ferramentas autônomas e operação de computador: OSWorld, Toolathlon, MCP Atlas, τ²-Bench.

Geral767 Código469 Matemática90 Raciocínio483 Agentes89 Conhecimento482 Multimodal452 Agentes de código68 Raciocínio científico96 Multilíngue12 Escrita95 Chat e assistentes95 Pesquisa99 Chamada de ferramentas45 Contexto longo116 Velocidade472 Custo-benefício471 Pesos abertos183

Modelos: 89

#	Modelo	Podium Score	OSWorld	Toolathlon	MCP Atlas	τ²-Bench Retail	Apex Agents	Velocidade
01	Kimi K3 Moonshot AIAbertoraciocínio	100.0	—	73.2%	84.2%	—	37.6%	37 t/s
02	Claude Fable 5 Anthropicraciocínio	96.2	—	—	—	98.5%	—	71 t/s
03	Claude 5.5 Sonnet Anthropicraciocínio	89.4	84.5%	—	—	92.5%	—	94 t/s
4	Gemini 4 Argon Googleraciocínioprévia	87.0	81%	—	—	90%	—	88 t/s
5	GPT-6.1 Sol OpenAIraciocínio	86.5	80%	—	—	89%	—	75 t/s
6	MiMo v2.6 Flash XiaomiAberto	84.0	74%	—	—	83%	—	118 t/s
7	Kimi K3.1 Pro Moonshot AIAbertoraciocínio	82.0	—	—	—	90%	—	45 t/s
8	Gemini 3.1 Pro Preview Googleraciocínioprévia	82.0	75%	—	—	86%	—	105 t/s
9	Mistral Large 4 Mistralraciocínio	81.0	72%	—	—	84%	—	92 t/s
10	NVIDIA Nemotron 3.5 Super-VL NVIDIAAberto	81.0	71%	—	—	80%	—	110 t/s
11	Claude 5.5 Haiku Anthropicraciocínio	78.5	68%	—	—	81%	—	145 t/s
12	Gemini 3.5 Flash Google	75.9	—	—	—	95.3%	—	267 t/s
13	Claude Opus 4.8 Anthropic	74.9	—	59.9%	82.2%	94.4%	—	50 t/s
14	Claude Opus 4.7 Anthropic	74.3	—	—	77.3%	—	—	80 t/s
15	Kimi K3.1 Highspeed Moonshot AIAbertoraciocínio	74.0	—	—	—	84%	—	130 t/s
16	Gemini 3.1 Flash Lite Preview Googleprévia	74.0	62%	—	—	75%	—	210 t/s
17	GPT-6 Astra OpenAIraciocínio	72.6	72.6%	—	—	88%	—	82 t/s
18	Deepgram Flux Independent	72.0	—	—	—	—	—	260 t/s
19	Qwen3.7 Max Alibaba	71.5	—	—	76.4%	94.7%	—	203 t/s
20	ElevenLabs Scribe v2 ElevenLabs	70.0	—	—	—	—	—	240 t/s
21	Simba 3.2 Independentraciocínio	70.0	—	—	—	78%	—	62 t/s
22	ElevenLabs v4 Turbo Independent	70.0	—	—	—	—	—	220 t/s
23	MiMo V2.5 Pro XiaomiAberto	69.0	—	—	—	94.2%	—	65 t/s
24	Cartesia Ink Whisper Cartesia	68.0	—	—	—	—	—	260 t/s
25	GLM-5.2 Z.aiAberto	66.9	—	48.2%	76.8%	99.1%	—	164 t/s
26	Eleven Multilingual v3 ElevenLabs	65.0	—	—	—	—	—	200 t/s
27	Qwen3.8 Flash Next AlibabaAberto	63.5	—	—	—	—	—	168 t/s
28	Gemini 3.1 Pro Google	62.6	—	—	69.2%	95.6%	33.5%	136 t/s
29	GLM-4.5 Air Zhipu AIAbertoraciocínio	62.0	—	—	—	76%	—	96 t/s
30	GPT-5.5 OpenAI	61.1	—	55.6%	75.3%	93.9%	—	60 t/s
31	GLM-5.3 Flash Z.aiAberto	61.0	—	—	—	—	—	182 t/s
32	Qwen3.7 Plus Alibaba	59.0	—	—	73.2%	—	—	54 t/s
33	Ling 3.1 Flash Ling	58.0	—	—	—	—	—	175 t/s
34	Granite 4.2 30B Instruct IBMAberto	58.0	—	—	—	—	—	78 t/s
35	Step Audio 2 StepFun	55.0	—	—	—	—	—	150 t/s
36	Qwen3.6 Plus Alibaba	52.6	—	39.8%	74.1%	97.7%	—	55 t/s
37	Voxtral 3 TTS Mistral	52.0	—	—	—	—	—	165 t/s
38	Muse Spark Meta	51.9	—	—	—	91.5%	—	80 t/s
39	Qwen Audio 3.1 TTS+ AlibabaAberto	50.0	—	—	—	—	—	140 t/s
40	DeepSeek V4 Pro DeepSeekAberto	49.7	—	51.8%	73.6%	91.2%	—	66 t/s
41	DeepSeek V4.1 Flash DeepSeekAbertoraciocínio	48.0	—	—	—	—	—	214 t/s
42	Qwen3 TTS Flash Alibaba	45.0	—	—	—	—	—	230 t/s
43	GPT-5.6 Luna OpenAI	43.3	—	53.4%	—	—	—	175 t/s
44	Claude Opus 4.6 Anthropic	43.1	72.7%	—	62.7%	84.8%	—	50 t/s
45	Kimi K2.7 Code Moonshot AIAberto	43.0	—	—	—	90.1%	—	39 t/s
46	Granite 4.2 8B Instruct IBMAberto	42.0	—	—	—	—	—	115 t/s
47	GPT-5.4 OpenAI	41.6	—	54.6%	67.2%	—	—	60 t/s
48	Agnes 3.0 Flash Agnesraciocínio	40.0	—	—	—	—	—	235 t/s
49	Agnes 2.5 Pro Beta Agnes	37.0	—	—	—	—	—	65 t/s
50	Hunyuan Hy3 Tencent	36.7	—	48.5%	79.1%	—	25.6%	69 t/s
51	DeepSeek V4 Flash Vision DeepSeek	36.0	—	—	—	—	—	120 t/s
52	G9v3-39A5B AI9StarsAberto	36.0	—	—	—	—	—	60 t/s
53	G9v3-3B AI9StarsAberto	36.0	—	—	—	—	—	60 t/s
54	GLM-5.3 Zhipu AI	36.0	—	—	—	—	—	90 t/s
55	Grok 4.6 (low) xAI	36.0	—	—	—	—	—	59 t/s
56	Grok 4.6 (medium) xAI	36.0	—	—	—	—	—	59 t/s
57	Grok 4.6 (xhigh) xAI	36.0	—	—	—	—	—	62 t/s
58	HyperNova 60B MultiverseAberto	36.0	—	—	—	—	—	367 t/s
59	LFM2.5-2.6B Liquid AIAberto	36.0	—	—	—	—	—	60 t/s
60	LFM2.5-8B-A1B Liquid AIAberto	36.0	—	—	—	—	—	338 t/s
61	LongCat 2.0 LongCatAberto	36.0	—	—	—	—	—	42 t/s
62	Magistral Small 1.2 MistralAberto	36.0	—	—	—	—	—	60 t/s
63	Qwen3.8 27B (xhigh) AlibabaAberto	36.0	—	—	—	—	—	53 t/s
64	Qwen3.8 27B (low) AlibabaAberto	36.0	—	—	—	—	—	64 t/s
65	Qwen3.8 27B (medium) AlibabaAberto	36.0	—	—	—	—	—	62 t/s
66	Qwen3.8 27B (Non-reasoning) AlibabaAbertoraciocínio	36.0	—	—	—	—	—	62 t/s
67	MiniMax M3 MiniMax	35.4	—	—	—	88.9%	—	93 t/s
68	DeepSeek V4 Flash DeepSeekAberto	35.3	—	47.8%	69%	—	—	113 t/s
69	GPT-5.6 Sol OpenAI	33.9	—	58%	—	85.1%	—	72 t/s
70	FLUX.2 Pro Black Forest Labs	32.0	—	—	—	—	—	30 t/s
71	GPT-5.6 Terra OpenAI	30.7	—	53.1%	—	86.3%	—	127 t/s
72	GPT-Image 2 OpenAI	30.0	—	—	—	—	—	25 t/s
73	FLUX.2 Max Independent	30.0	—	—	—	—	—	35 t/s
74	Grok Imagine Quality xAI	28.0	—	—	—	—	—	34 t/s
75	FLUX.2 Dev Black Forest LabsAberto	28.0	—	—	—	—	—	28 t/s
76	Kimi K2.6 Moonshot AIAberto	26.3	—	50%	—	—	27.9%	40 t/s
77	Kling 3.0 Pro Kwai	25.0	—	—	—	—	—	25 t/s
78	Gemini 3 Pro Google	24.1	—	—	—	87.1%	—	120 t/s
79	LTX-2.3 Pro Lightricks	22.0	—	—	—	—	—	18 t/s
80	FLUX.2 Klein 9B Black Forest LabsAberto	22.0	—	—	—	—	—	55 t/s
81	Wan 2.2 A14B AlibabaAberto	20.0	—	—	—	—	—	15 t/s
82	LTX-2.3 Fast Lightricks	20.0	—	—	—	—	—	60 t/s
83	FLUX.2 Klein 4B Black Forest LabsAberto	18.0	—	—	—	—	—	90 t/s
84	GPT-5.2 OpenAI	17.4	—	46.3%	60.6%	—	—	90 t/s
85	GPT-5.3 Codex OpenAI	17.1	—	—	—	86%	—	118 t/s
86	Gemini 3 Flash Google	15.9	—	49.4%	57.4%	—	—	120 t/s
87	Claude Sonnet 4.6 Anthropic	7.3	72.5%	—	61.3%	—	—	50 t/s
88	Qwen3.5 397B-A17B AlibabaAberto	0.0	—	38.3%	—	—	—	71 t/s
89	Nemotron 3 Ultra 550B NVIDIAAberto	0.0	—	—	—	83.3%	—	142 t/s

O Podium Score é uma mistura ponderada de cinco leaderboards. Saiba mais na metodologia.

Model Name

Provider

Podium Score0.0

Context128K

Price (1M out)$0/M

Speed0 t/s

Architecture & Highlights

Model overview and workload profile.

Quickstart cURL

curl https://api.openrouter.ai/api/v1/chat/completions ...