Ranking 2026

Melhores LLMs Gratuitos

297 modelos de IA sem custo — open-source e planos gratuitos de API. Ordenados por AA Intelligence Index (Artificial Analysis, atualizado a cada 6h).

297
modelos gratuitos
114
open-source
183
API gratuita

Top 5 LLMs Gratuitos

#1open-source
Qwen: Qwen3.8 Max
AlibabaAlibaba
Score AA 58.1
Velocidade 82 tok/s
#2open-source
DeepSeek V4 Flash
DeepSeekDeepSeek
Score AA 51.8
Velocidade 136 tok/s
#3open-source
DeepSeek V4 Pro
DeepSeekDeepSeek
Score AA 45.3
ELO 1457
Velocidade 77 tok/s
#4open-source
MoonshotAI: Kimi K2.6
MoonshotAIMoonshotAI
Score AA 45.1
ELO 1461
Velocidade 44 tok/s
#5
Motif 3 (Beta)
Motif Technologies
Score AA 44.9
0

Lista Completa de LLMs Gratuitos

#ModeloScore AA
1Qwen: Qwen3.8 Max58.1
2DeepSeek V4 Flash51.8
3DeepSeek V4 Pro45.3
4MoonshotAI: Kimi K2.645.1
5Motif 3 (Beta)44.9
6Muse Spark43.5
7Xiaomi: MiMo-V2-Pro41.4
8Qwen: Qwen3.6 Plus40.5
9JT-4.1 Flash 236B A21B39.8
10GLM-5-Turbo39.1
11MiniMax: MiniMax M2.738.9
12Qwen: Qwen3.6 27B37.7
13MiMo-V2-Omni-032737.3
14MoonshotAI: Kimi K2.536.0
15Xiaomi: MiMo-V2-Omni35.9
16GLM 5V Turbo (Reasoning)35.3
17Qwen: Qwen3.5-27B34.6
18MiniMax: MiniMax M2.534.5
19Qwen: Qwen3.5 397B A17B34.3
20MiMo-V2-Flash (Feb 2026)34.0
21Qwen: Qwen3.5-122B-A10B32.8
22Qwen: Qwen3 Max Thinking32.5
23Qwen: Qwen3.6 35B A3B32.1
24MiniMax: MiniMax M2.132.1
25G9v3-39A5B31.4
26Grok 4.1 Fast31.3
27DeepSeek V3.1 Terminus30.6
28Mistral: Mistral Medium 3.530.4
29Qwen: Qwen3.5-35B-A3B29.9
30JT-35B-Flash29.0
31KAT-Coder-Pro V128.9
32Claude 3.7 Sonnet (thinking)27.6
33Doubao Seed Code26.5
34DeepSeek V3.2 Exp25.9
35DeepSeek V3.225.1
36Qwen: Qwen3 Max24.5
37Ling 3.0 Tiny24.5
38Gemini 2.5 Flash Preview (Sep '25) (Reasoning)24.2
39MoonshotAI: Kimi K2 090524.0
40Gemini 2.5 Pro Preview (Mar' 25)23.4
41Command A+22.8
42DeepSeek V3.2 Speciale22.6
43K-EXAONE (Reasoning)22.4
44ERNIE 5.0 Thinking Preview22.3
45Grok Code Fast 122.0
46Apriel-v1.5-15B-Thinker21.6
47DeepSeek V3.121.4
48Qwen: Qwen3 Coder Next21.3
49Apriel-v1.6-15B-Thinker20.8
50EXAONE 4.5 33B20.5
51North Mini Code20.2
52Qwen: Qwen3 235B A22B Instruct 250719.9
53Qwen: Qwen3 235B A22B Thinking 250719.9
54Mistral: Mistral Small 419.7
55GLM-4.5 (Reasoning)19.7
56Devstral 219.2
57JT-MINI18.8
58Arcee AI: Trinity Large Thinking18.6
59DeepSeek: R118.6
60Nemotron Cascade 2 30B A3B18.0
61Sonar Reasoning Pro18.0
62Devstral Small 217.7
63LongCat Flash Lite17.4
64K2 Think V217.2
65HyperCLOVA X SEED Think (32B)17.2
66Mi:dm K 2.5 Pro16.6
67G9v3-3B16.5
68Mi:dm K 2.5 Pro Preview16.4
69Ring-1T16.3
70INTELLECT-315.7
71GPT-5 (ChatGPT)15.4
72Solar Open 100B (Reasoning)15.2
73Mistral: Mistral Medium 3.114.7
74Llama 4 Maverick14.5
75MiniMax M1 40k14.5
76Qwen: Qwen3 VL 235B A22B Instruct14.4
77DeepSeek V314.2
78K2-V2 (high)14.2
79K2-V2 (medium)14.2
80o1-mini14.0
81Qwen: Qwen3 Next 80B A3B Instruct13.8
82DiffusionGemma 26B A4B13.8
83Qwen: Qwen3 Coder 30B A3B Instruct13.6
84GPT-4.5 (Preview)13.6
85Tri-21B-think Preview13.6
86Gemini 2.0 Flash Thinking Experimental (Jan '25)13.3
87Motif-2-12.7B-Reasoning12.8
88Ling-1T12.7
89Magistral Medium 112.5
90Mistral: Mistral Medium 312.5
91Solar Pro 2 (Preview) (Reasoning)12.5
92Claude 3.5 Haiku12.4
93Mistral: Devstral Medium12.4
94GPT-4o (March 2025, chatgpt-4o-latest)12.3
95Tri-21B-Think12.3
96Gemini 2.0 Flash12.2
97Llama 3.3 Nemotron Super 49B v1 (Reasoning)12.2
98Qwen3 4B 2507 (Reasoning)11.9
99MiniCPM5-1B (Non-reasoning)11.9
100Gemini 2.0 Pro Experimental (Feb '25)11.8
101Devstral Small (May '25)11.8
102Gemini 2.5 Flash Preview (Non-reasoning)11.6
103Sonar Reasoning11.6
104Nanbeige4.1-3B11.2
105Qwen: Qwen3 VL 32B Instruct11.0
106DeepSeek: R1 Distill Qwen 32B11.0
107Gemini 2.0 Flash (experimental)10.6
108Magistral Small 110.6
109EXAONE 4.0 32B (Reasoning)10.5
110DeepSeek R1 0528 Qwen3 8B10.3
111Llama 4 Scout10.3
112Qwen2.5 Max10.1
113Qwen: Qwen3 VL 30B A3B Instruct9.9
114Gemini 1.5 Pro (Sep '24)9.9
115Solar Pro 2 (Preview) (Non-reasoning)9.9
116R1 Distill Llama 70B9.8
117Gemma 4 E2B (Reasoning)9.8
118DeepSeek R1 Distill Qwen 14B9.7
119Falcon-H1R-7B9.7
120Qwen2.5 72B Instruct9.4
121Sonar9.4
122Llama 3.3 70B Instruct9.3
123Step3 VL 10B9.3
124QwQ 32B-Preview9.1
125Devstral Small (Jul '25)9.1
126Baidu: ERNIE 4.5 300B A47B 8.9
127Granite 4.1 30B8.9
128NVIDIA Nemotron 3 Nano 4B8.8
129Solar Pro 2 (Non-reasoning)8.8
130Gemini 2.0 Flash-Lite (Feb '25)8.6
131Gemini 2.0 Flash-Lite (Preview)8.4
132Kimi Linear 48B A3B Instruct8.4
133Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)8.4
134Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)8.3
135Qwen: Qwen3 VL 8B Instruct8.2
136Qwen3 4B (Reasoning)8.2
137Llama 3.1 Tulu3 405B8.1
138LFM2.5-8B-A1B8.1
139GPT-4o (ChatGPT)8.1
140Mistral: Pixtral Large 24118.0
141Olmo 3.1 32B Think7.9
142Gemini 1.5 Flash (Sep '24)7.8
143Llama 3.1 8B Instruct7.8
144Grok 2 (Dec '24)7.8
145Qwen3 VL 4B (Reasoning)7.7
146Qwen3.5 2B (Reasoning)7.5
147Gemma 3 27B7.4
148Llama 3.1 Nemotron 70B Instruct7.4
149Grok Beta7.3
150Qwen2.5 Instruct 32B7.2
151Qwen2.5 Coder 32B Instruct6.9
152Qwen3 4B 2507 Instruct6.9
153Qwen: Qwen3 30B A3B Instruct 25076.6
154Qwen: Qwen3 30B A3B Thinking 25076.6
155Qwen3 4B (Non-reasoning)6.5
156DeepSeek-V2.5 (Dec '24)6.5
157Llama 3.1 70B Instruct6.5
158DeepSeek-V2.56.4
159Gemini 2.0 Flash Thinking Experimental (Dec '24)6.4
160Olmo 3.1 32B Instruct6.2
161DeepSeek R1 Distill Llama 8B6.2
162Gemma 4 E2B (Non-reasoning)6.2
163Phi-4 Mini Instruct6.2
164Mistral: Saba6.2
165Olmo 3 32B Think6.1
166Gemini 1.5 Pro (May '24)6.1
167Qwen: Qwen-Turbo6.0
168Llama 3.2 Instruct 90B (Vision)6.0
169R1 17766.0
170Reka Flash 36.0
171Grok-15.8
172Qwen2 Instruct 72B5.7
173EXAONE 4.0 32B (Non-reasoning)5.7
174Gemma 3 12B5.5
175Gemini 1.5 Flash-8B5.2
176AI21: Jamba Large 1.75.0
177Qwen3.5 0.8B (Reasoning)5.0
178DeepHermes 3 - Mistral 24B Preview (Non-reasoning)5.0
179OLMo 2 32B4.7
180DeepSeek-Coder-V24.7
181Granite 4.1 3B4.7
182Gemini 1.5 Flash (May '24)4.6
183LFM2-24B-A2B4.6
184Microsoft: Phi 44.6
185Gemini 1.0 Ultra4.3
186Phi-3 Mini Instruct 3.8B4.3
187MiniCPM-V 4.6 1.3B4.3
188Gemma 3n E4B Instruct Preview (May '25)4.2
189Phi-4 Multimodal Instruct4.2
190Qwen2.5 Coder Instruct 7B 4.1
191Mistral Large4.1
192Mixtral 8x22B Instruct4.0
193Mistral: Mixtral 8x22B Instruct4.0
194Llama 3.2 3B Instruct3.9
195Jamba Reasoning 3B3.8
196Qwen1.5 Chat 110B3.7
197Qwen3 VL 4B Instruct3.7
198Olmo 3 7B Think3.6
199Qwen3.5 0.8B (Non-reasoning)3.5
200OLMo 2 7B3.5
201Claude 2.13.5
202Molmo 7B-D3.4
203Ling-mini-2.03.4
204Claude 2.03.3
205DeepSeek R1 Distill Qwen 1.5B3.3
206DeepSeek-V2-Chat3.3
207Llama 3 70B Instruct3.1
208Qwen Chat 72B3.0
209LFM 40B3.0
210Llama 3.2 11B Vision Instruct3.0
211Arctic Instruct3.0
212PALM-22.8
213DeepSeek Coder V2 Lite Instruct2.7
214Gemini 1.0 Pro2.7
215DBRX Instruct2.6
216DeepSeek LLM 67B Chat (V1)2.6
217Llama 2 Chat 13B2.6
218Llama 2 Chat 70B2.6
219OpenChat 3.5 (1210)2.6
220Sarvam M (Reasoning)2.6
221Exaone 4.0 1.2B (Non-reasoning)2.5
222Jamba 1.7 Mini2.3
223LFM2 2.6B2.3
224LFM2.5-1.2B-Instruct2.3
225LFM2.5-1.2B-Thinking2.3
226Qwen3 1.7B (Reasoning)2.2
227Granite 4.0 H 1B2.2
228Gemma 3 270M2.0
229Granite 4.0 Micro2.0
230Mistral: Mixtral 8x7B Instruct2.0
231DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning)1.9
232Qwen Chat 14B1.7
233Claude Instant1.7
234Llama 65B1.7
235Molmo2-8B1.6
236Granite 4.0 1B1.6
237LFM2 8B A1B1.3
238Qwen3 1.7B (Non-reasoning)1.1
239Qwen3 0.6B (Non-reasoning)1.0
240Qwen3 0.6B (Reasoning)1.0
241Tiny Aya Global1.0
242Gemma 3 1B Instruct1.0
243Gemma 3 4B1.0
244Gemma 3n E2B Instruct1.0
245Granite 4.0 350M1.0
246Granite 4.0 H 350M1.0
247LFM2 1.2B1.0
248LFM2.5-VL-1.6B1.0
249Llama 3 8B Instruct1.0
250Llama 3.2 1B Instruct1.0
251AlfredPros: CodeLLaMa 7B Instruct Solidity—
252Qwen: Qwen2.5 7B Instruct—
253Qwen: Qwen2.5 VL 72B Instruct—
254Wan 2.1—
255Arcee AI: Trinity Mini—
256Baidu: ERNIE 4.5 21B A3B Thinking—
257Baidu: ERNIE 4.5 VL 28B A3B—
258Baidu: ERNIE 4.5 VL 424B A47B —
259ByteDance: UI-TARS 7B —
260DeepSeek: DeepSeek V4 Flash 0423—
261EssentialAI: Rnj 1 Instruct—
262Goliath 120B—
263Gemini 3 Deep Think—
264Ideogram 4.0—
265Magnum v4 72B—
266Llama Guard 3 8B—
267Llama Guard 4 12B—
268WizardLM-2 8x22B—
269MiniMax: MiniMax-01—
270Mistral: Mistral Nemo—
271Mistral: Voxtral Small 24B 2507—
272MoonshotAI: Kimi K2 0711—
273MythoMax 13B—
274Nex AGI: DeepSeek V3.1 Nex N1—
275Nous: Hermes 3 405B Instruct—
276Nous: Hermes 3 70B Instruct—
277Nous: Hermes 4 405B—
278Nous: Hermes 4 70B—
279NousResearch: Hermes 2 Pro - Llama-3 8B—
280GPT-3.5 Turbo (0613)—
281GPT-4o mini Realtime (Dec '24)—
282GPT-4o Realtime (Dec '24)—
283Reka Edge—
284ReMM SLERP 13B—
285Sao10K: Llama 3 8B Lunaris—
286Sao10k: Llama 3 Euryale 70B v2.1—
287Sao10K: Llama 3.1 70B Hanami x1—
288Sao10K: Llama 3.1 Euryale 70B v2.2—
289Sao10K: Llama 3.3 Euryale 70B—
290Stable Diffusion 3.5 Large—
291Stable Video Diffusion 3D—
292Tencent: Hunyuan A13B Instruct—
293TheDrummer: Cydonia 24B V4.1—
294TheDrummer: Rocinante 12B—
295TheDrummer: Skyfall 36B V2—
296TheDrummer: UnslopNemo 12B—
297Tongyi DeepResearch 30B A3B—

Como Acessar LLMs Gratuitos

Via Interface Web (sem código)

  • ChatGPT — GPT-4o mini grátis em chat.openai.com
  • Claude.ai — Claude 3.5 Haiku grátis (limite diário)
  • Gemini — Gemini 1.5 Flash grátis em gemini.google.com
  • Mistral Le Chat — Mistral Large grátis
  • Groq — Llama 3, Mistral via API rápida grátis

Rodar Localmente (open-source)

  • Ollama — instale e execute ollama run llama3
  • LM Studio — interface gráfica para modelos locais
  • Jan.ai — app desktop com interface amigável
  • Together.ai — inferência cloud com cota gratuita generosa
  • Replicate — pay-per-use, mas com créditos iniciais

Open-Source vs. API Gratuita: Qual Escolher?

CritérioOpen-SourceAPI Gratuita
Privacidade dos dadosTotal (roda local)Compartilhados com provedor
Custo de infraestruturaGPU própria ou cloudZero (dentro do limite)
Limite de usoIlimitadoRate limit diário/mensal
Facilidade de setupMédio (requer técnica)Imediato
Customização/fine-tuningTotalLimitada ou indisponível
Qualidade máxima disponívelLlama 3.1 405BGPT-4o, Gemini Ultra

Perguntas Frequentes

Qual o melhor LLM gratuito disponível hoje?▾
Com base no AA Intelligence Index, o melhor modelo gratuito atualmente é o Qwen: Qwen3.8 Max (score 58.1), que pode ser acessado gratuitamente. Em seguida aparecem DeepSeek V4 Flash e DeepSeek V4 Pro.
Qual a diferença entre LLM gratuito e open-source?▾
LLM gratuito significa que você pode usá-lo sem pagar — via interface web (ChatGPT free, Claude.ai, Gemini) ou via API com tier gratuito. Open-source vai além: o código e os pesos são públicos, permitindo rodar localmente, modificar e usar sem limites ou privacidade comprometida. Exemplos open-source: Llama 3, Mistral, Qwen. Exemplos gratuitos mas não open-source: GPT-4o-mini (API grátis com limite), Gemini Flash.
Como rodar um LLM open-source localmente?▾
A forma mais fácil é usar o Ollama (ollama.ai): instale, execute "ollama run llama3" e o modelo roda no seu computador. Para modelos maiores (70B+), é necessário GPU com pelo menos 40GB VRAM ou usar serviços como Groq, Together.ai e Replicate — que oferecem inferência gratuita com cota mensal.
LLMs gratuitos têm qualidade suficiente para uso profissional?▾
Sim, para muitos casos de uso. Modelos como Llama 3.1 70B, Mistral Large e Qwen 2.5 72B atingem performance comparável ou superior a GPT-3.5 e competem com GPT-4 em tarefas específicas. Para uso profissional de alto volume (produção, integração via API, SLA), a API paga geralmente oferece melhor confiabilidade e suporte.

Conteúdo relacionado:

Benchmark Open-SourceFerramentas GratuitasRanking Custo-BenefícioMelhores LLMs 2026Benchmark CompletoComo Escolher um LLM