- Atualizado em: America/Sao Paulo - 26/09/2026 15:27
- Source: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings).
Matriz de Custo-Benefício dos Modelos
Coding Index (Y) × Preço por milhão de tokens (X, invertido). Quanto mais ao topo e à direita, melhor a relação capacidade/preço. A matriz usa valores factuais sem nenhum cálculo. O corte de 60 pontos vs 4$ é empirico.
Superior Esquerdo: Alta Capacidade / Alto Custo (Top de linha / Premium)
Superior Direito: 🏆 Melhor Custo-Benefício (Alta Capacidade + Baixo Custo)
Inferior Esquerdo: Baixa Eficiência (Custo Alto para a capacidade oferecida)
Inferior Direito: Econômico (Entrada / Baixo Custo)
| Rank | Modelo | Slug | Coding | Preço | Gasto | Eficiencia |
|---|---|---|---|---|---|---|
| 1 | GLM 5.3 Flash | z-ai/glm-5.3-flash-20260826 |
71.5 | $0.650 | 0.009 | 68.250 |
| 2 | DeepSeek V4 Flash 0731 (Reasoning, Max Effort) | deepseek/deepseek-v4-flash-20260731 |
69.1 | $0.420 | 0.006 | 67.000 |
| 3 | GPT-5.6 Luna (max) | openai/gpt-5.6-luna-20260709 |
71.4 | $2.800 | 0.039 | 57.400 |
| 4 | Ling-3.0-flash-VL | inclusionai/ling-3.0-flash-vl-20260910 |
57.0 | $0.083 | 0.001 | 56.585 |
| 5 | Hy3 | tencent/hy3-preview-20260421 |
58.8 | $0.780 | 0.013 | 54.900 |
| 6 | Ling-3.0-flash-Fin | inclusionai/ling-3.0-flash-fin-20260827 |
55.6 | $0.240 | 0.004 | 54.400 |
| 7 | GPT-5.4 nano (xhigh) | openai/gpt-5.4-nano-20260317 |
56.1 | $0.725 | 0.013 | 52.475 |
| 8 | MiMo-V2.5-Pro | xiaomi/mimo-v2.5-pro-20260422 |
60.2 | $1.566 | 0.026 | 52.370 |
| 9 | Solar Pro 4 | upstage/solar-pro4-20260810 |
52.7 | $0.450 | 0.009 | 50.450 |
| 10 | Ling 3.0 Flash | inclusionai/ling-3.0-flash-20260723 |
50.6 | $0.084 | 0.002 | 50.180 |
| 11 | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) | deepseek/deepseek-v4-pro-20260813 |
68.8 | $3.750 | 0.055 | 50.050 |
| 12 | DeepSeek V4 Flash 0420 (Reasoning, Max Effort) | deepseek/deepseek-v4-flash-20260423 |
56.2 | $1.310 | 0.023 | 49.650 |
| 13 | Qwen3.7 Plus | qwen/qwen3.7-plus-20260602 |
55.9 | $1.600 | 0.029 | 47.900 |
| 14 | Qwen3.8 27B (xhigh) | qwen/qwen3.8-27b-20260814 |
68.1 | $4.491 | 0.066 | 45.645 |
| 15 | MiniMax-M2.7 | minimax/minimax-m2.7-20260318 |
52.6 | $1.500 | 0.029 | 45.100 |
| 16 | MiMo-V2.5 | xiaomi/mimo-v2.5-20260422 |
56.8 | $2.400 | 0.042 | 44.800 |
| 17 | Muse Spark 1.2 (xhigh) | meta/muse-spark-1.2-20260805 |
72.2 | $5.500 | 0.076 | 44.700 |
| 18 | Muse Spark 1.1 (xhigh) | meta/muse-spark-1.1-20260709 |
71.3 | $5.500 | 0.077 | 43.800 |
| 19 | MiniMax-M3 | minimax/minimax-m3-20260531 |
58.6 | $3.000 | 0.051 | 43.600 |
| 20 | Qwen3.6 Plus | qwen/qwen3.6-plus-04-02 |
54.5 | $2.275 | 0.042 | 43.125 |
| 21 | GLM-5.3 (max) | z-ai/glm-5.3-20260816 |
74.8 | $6.380 | 0.085 | 42.900 |
| 22 | Gemma 4 31B (Reasoning) | google/gemma-4-31b-it-20260402 |
43.4 | $0.380 | 0.009 | 41.500 |
| 23 | DeepSeek V4 Pro 0424 (Reasoning, Max Effort) | deepseek/deepseek-v4-pro-20260423 |
59.4 | $3.700 | 0.062 | 40.900 |
| 24 | Kimi K2.6 | moonshotai/kimi-k2.6-20260420 |
61.8 | $4.455 | 0.072 | 39.525 |
| 25 | Qwen3.6 27B (Reasoning) | qwen/qwen3.6-27b-20260422 |
53.7 | $3.020 | 0.056 | 38.600 |
| 26 | LongCat 2.0 | meituan/longcat-2.0-20260720 |
45.3 | $1.500 | 0.033 | 37.800 |
| 27 | DeepSeek V3.1 Terminus (Reasoning) | deepseek/deepseek-v3.1-terminus |
43.5 | $1.371 | 0.032 | 36.645 |
| 28 | Qwen3.7 Max | qwen/qwen3.7-max-20260520 |
66.0 | $5.900 | 0.089 | 36.500 |
| 29 | Qwen3.6 35B A3B (Reasoning) | qwen/qwen3.6-35b-a3b-20260415 |
41.9 | $1.100 | 0.026 | 36.400 |
| 30 | Qwen3.8 Max (0902) | qwen/qwen3.8-max-20260902 |
76.2 | $8.000 | 0.105 | 36.200 |
| 31 | Kimi K2.7 Code | moonshotai/kimi-k2.7-code-20260612 |
60.8 | $4.950 | 0.081 | 36.050 |
| 32 | Gemini 3.8 Flash (high) | google/gemini-3.8-flash-20260902 |
76.3 | $8.100 | 0.106 | 35.800 |
| 33 | Gemini 3.7 Flash (high) | google/gemini-3.7-flash-20260813 |
76.1 | $8.100 | 0.106 | 35.600 |
| 34 | GLM-4.6 (Reasoning) | z-ai/glm-4.6 |
45.8 | $2.180 | 0.048 | 34.900 |
| 35 | Nemotron 3 Ultra 550B A55B (Reasoning) | nvidia/nemotron-3-ultra-550b-a55b-20260604 |
49.3 | $3.000 | 0.061 | 34.300 |
| 36 | Qwen3.5 122B A10B (Reasoning) | qwen/qwen3.5-122b-a10b-20260224 |
45.7 | $2.690 | 0.059 | 32.250 |
| 37 | Qwen3.8 2.4T A95B | qwen/qwen3.8-2.4t-a95b-20260812 |
71.9 | $8.000 | 0.111 | 31.900 |
| 38 | GLM-4.7 (Reasoning) | z-ai/glm-4.7-20251222 |
45.3 | $3.200 | 0.071 | 29.300 |
| 39 | Kimi K2.5 (Reasoning) | moonshotai/kimi-k2.5-0127 |
46.8 | $3.600 | 0.077 | 28.800 |
| 40 | Gemini 3.6 Flash (high) | google/gemini-3.6-flash-20260721 |
69.2 | $8.100 | 0.117 | 28.700 |
| 41 | GLM-5.1 (Reasoning) | z-ai/glm-5.1-20260406 |
55.8 | $5.510 | 0.099 | 28.250 |
| 42 | Qwen3.5 397B A17B (Reasoning) | qwen/qwen3.5-397b-a17b-20260216 |
48.2 | $4.050 | 0.084 | 27.950 |
| 43 | Grok 4.3 (high) | x-ai/grok-4.3-20260430 |
42.2 | $3.000 | 0.071 | 27.200 |
| 44 | Mistral Medium 3.5 | mistralai/mistral-medium-3.5-20260430 |
46.9 | $4.500 | 0.096 | 24.400 |
| 45 | Gemini 3.5 Flash-Lite | google/gemini-3.5-flash-lite-20260721 |
49.3 | $5.040 | 0.102 | 24.100 |
| 46 | Grok Build 0.1 0616 | x-ai/grok-build-0.1-20260520 |
51.5 | $6.000 | 0.117 | 21.500 |
| 47 | GLM-5.2 (max) | z-ai/glm-5.2-20260616 |
68.8 | $10.250 | 0.149 | 17.550 |
| 48 | Claude 4.5 Haiku (Reasoning) | anthropic/claude-4.5-haiku-20251001 |
43.9 | $6.600 | 0.150 | 10.900 |
| 49 | Kimi K3 (max) | moonshotai/kimi-k3-20260715 |
76.2 | $13.680 | 0.180 | 7.800 |
| 50 | DeepSeek V3.2 (Reasoning) | deepseek/deepseek-v3.2-20251201 |
44.2 | $7.500 | 0.170 | 6.700 |
| 51 | Claude Sonnet 5 (Adaptive Reasoning, Max Effort) | anthropic/claude-sonnet-5-20260630 |
71.5 | $13.200 | 0.185 | 5.500 |
| 52 | GPT-5.4 mini (xhigh) | openai/gpt-5.4-mini-20260317 |
56.1 | $10.500 | 0.187 | 3.600 |
| 53 | Grok 4.6 (high) | x-ai/grok-4.6-20260810 |
76.8 | $16.000 | 0.208 | -3.200 |
| 54 | Grok 4.5 (high) | x-ai/grok-4.5-20260708 |
72.4 | $16.000 | 0.221 | -7.600 |
| 55 | GPT-5.5 (xhigh) | openai/gpt-5.5-20260423 |
74.9 | $17.500 | 0.234 | -12.600 |
| 56 | Gemini 3.5 Flash (high) | google/gemini-3.5-flash-20260519 |
70.1 | $18.900 | 0.270 | -24.400 |
| 57 | Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort) | anthropic/claude-4.6-sonnet-20260217 |
63.0 | $18.000 | 0.286 | -27.000 |
| 58 | Claude 4.5 Sonnet (Reasoning) | anthropic/claude-4.5-sonnet-20250929 |
52.1 | $18.000 | 0.345 | -37.900 |
| 59 | GPT-5.6 Sol (xhigh) | openai/gpt-5.6-sol-20260709 |
78.3 | $24.000 | 0.307 | -41.700 |
| 60 | Gemini 3.1 Pro Preview | google/gemini-3.1-pro-preview-20260219 |
68.8 | $25.200 | 0.366 | -57.200 |
| 61 | GPT-5.1 (high) | openai/gpt-5.1-20251113 |
49.4 | $22.500 | 0.455 | -63.100 |
| 62 | GPT-5.6 Terra (max) | openai/gpt-5.6-terra-20260709 |
76.7 | $28.000 | 0.365 | -63.300 |
| 63 | Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) | anthropic/claude-fable-5.1-20260831 |
81.6 | $30.000 | 0.368 | -68.400 |
| 64 | Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | anthropic/claude-4.7-opus-20260416 |
73.6 | $30.000 | 0.408 | -76.400 |
| 65 | Claude Opus 5 (Adaptive Reasoning, Max Effort) | anthropic/claude-opus-5-20260723 |
78.0 | $33.000 | 0.423 | -87.000 |
| 66 | GPT-5.4 (xhigh) | openai/gpt-5.4-20260305 |
71.1 | $35.000 | 0.492 | -103.900 |
| 67 | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | anthropic/claude-5-fable-20260609 |
76.5 | $60.000 | 0.784 | -223.500 |
| 68 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | anthropic/claude-4.8-opus-20260528 |
74.3 | $60.000 | 0.808 | -225.700 |
| 69 | GPT-6 Astra (max) | openai/gpt-6-astra-20260903 |
76.9 | $66.000 | 0.858 | -253.100 |
Aqui listamos as melhores IAs para código, ranqueadas por critérios internos como eficiência e custo — priorizando modelos com bom desempenho em coding e melhor relação preço por ponto de capacidade.
Só entram modelos com índice de coding ≥ 40.0. O ranking usa a eficiência:
eficiencia = coding − 5 × preço
+$1.00 de preço vale +5 pontos de coding. Em caso de empate no score, vence o menor gasto (preço / coding).
- Coding: índice de coding da Artificial Analysis via OpenRouter.
- Preço: soma de input + output por milhão de tokens.
- Gasto: preço dividido pelo índice de coding (desempate; menor = melhor).
- Eficiência: a fórmula acima; maior valor = melhor posição.
Obs: O cálculo também é empírico.