开放权重 AI 模型：评测与价格
权重是自行运行模型所需的文件，还需要合适的软件和硬件。比较可下载权重的模型。开放权重不代表许可证一定开源、商业使用不受限制或托管免费。这里的 API 价格对应已报告的托管服务，部署前请查看各模型许可证。
应该选哪个？
单一基准测试无法覆盖所有用途。请用自己的提示词测试，比较准确性、响应时间、输入输出支持和实际账单。
AI 模型
| 模型 | 智能得分 | 基准测试费用（美元） | 输入价格 / 百万 token（美元） | 输出价格 / 百万 token（美元） | 输出速度 |
|---|---|---|---|---|---|
| A.X-K2SK Telecom | 21.50 | - | - | - | - |
| Apertus 70B InstructSwiss AI Initiative | 5.10 | - | - | - | - |
| Apertus 8B InstructSwiss AI Initiative | 4.80 | - | 0.10 | 0.20 | 127 托克/秒 |
| Apriel-v1.5-15B-ThinkerServiceNow | 13.80 | - | 0.00 | 0.00 | - |
| Apriel-v1.6-15B-ThinkerServiceNow | 13.40 | - | - | - | - |
| Arctic InstructSnowflake | 5.40 | - | - | - | - |
| Beam 1.0Reflection | 30.30 | $319 | 0.40 | 1.20 | 340 托克/秒 |
| Cogito v2.1 (Reasoning)Deep Cogito | - | - | - | - | - |
| Command ACohere | 7.00 | - | 2.50 | 10.00 | 40 托克/秒 |
| Command-R (Mar '24)Cohere | 5.00 | - | - | - | - |
| Command-R+ (Apr '24)Cohere | 5.30 | - | - | - | - |
| DBRX InstructDatabricks | 5.30 | - | - | - | - |
| DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning)Nous Research | 5.10 | - | - | - | - |
| DeepHermes 3 - Mistral 24B Preview (Non-reasoning)Nous Research | 6.10 | - | - | - | - |
| DeepSeek Coder V2 Lite InstructDeepSeek | 5.30 | - | - | - | - |
| DeepSeek LLM 67B Chat (V1)DeepSeek | 5.30 | - | - | - | - |
| DeepSeek R1 (Jan '25)DeepSeek | 11.40 | $461 | 2.00 | 4.00 | 149 托克/秒 |
| DeepSeek R1 0528 (May '25)DeepSeek | 13.10 | - | 1.85 | 2.75 | 33 托克/秒 |
| DeepSeek R1 0528 Qwen3 8BDeepSeek | 8.10 | - | - | - | - |
| DeepSeek R1 Distill Llama 70BDeepSeek | 7.90 | - | 0.70 | 1.10 | - |
| DeepSeek R1 Distill Llama 8BDeepSeek | 6.50 | - | - | - | - |
| DeepSeek R1 Distill Qwen 1.5BDeepSeek | 5.50 | - | - | - | - |
| DeepSeek R1 Distill Qwen 14BDeepSeek | 7.80 | - | - | - | - |
| DeepSeek R1 Distill Qwen 32BDeepSeek | 8.40 | - | - | - | - |
| DeepSeek V3 (Dec '24)DeepSeek | 8.50 | $25.4 | 0.32 | 0.89 | 46 托克/秒 |
| DeepSeek V3 0324DeepSeek | 9.70 | $19.9 | 0.24 | 0.90 | 117 托克/秒 |
| DeepSeek V3.1 (Non-reasoning)DeepSeek | 13.70 | - | 0.56 | 1.68 | 97 托克/秒 |
| DeepSeek V3.1 (Reasoning)DeepSeek | 13.50 | - | 0.58 | 1.68 | 120 托克/秒 |
| DeepSeek V3.1 Terminus (Non-reasoning)DeepSeek | 13.90 | - | 0.27 | 1.00 | 74 托克/秒 |
| DeepSeek V3.1 Terminus (Reasoning)DeepSeek | 14.80 | - | 1.64 | 2.75 | - |
| DeepSeek V3.2 (Non-reasoning)DeepSeek | 16.00 | - | 0.28 | 0.42 | - |
| DeepSeek V3.2 (Reasoning)DeepSeek | 21.50 | - | 0.28 | 0.42 | - |
| DeepSeek V3.2 Exp (Non-reasoning)DeepSeek | 13.90 | - | 0.28 | 0.42 | - |
| DeepSeek V3.2 Exp (Reasoning)DeepSeek | 16.60 | - | 0.28 | 0.42 | - |
| DeepSeek V3.2 SpecialeDeepSeek | 14.50 | - | - | - | - |
| DeepSeek V4 Flash 0420 (High)DeepSeek | 24.40 | $121 | 0.11 | 0.23 | - |
| DeepSeek V4 Flash 0420 (Max)DeepSeek | 24.20 | $149 | 0.11 | 0.23 | 53 托克/秒 |
| DeepSeek V4 Flash 0420 (Non-reasoning)DeepSeek | 18.90 | - | 0.09 | 0.19 | - |
| DeepSeek V4 Flash 0731 (Max)DeepSeek | 34.30 | $200 | 0.14 | 0.28 | 150 托克/秒 |
| DeepSeek V4 Pro 0424 (High)DeepSeek | 30.10 | - | 0.43 | 0.87 | 101 托克/秒 |
| DeepSeek V4 Pro 0424 (Max)DeepSeek | 30.40 | $277 | 0.43 | 0.87 | 104 托克/秒 |
| DeepSeek V4 Pro 0424 (Non-reasoning)DeepSeek | 20.80 | - | 0.43 | 0.87 | 94 托克/秒 |
| DeepSeek V4 Pro 0813 (Max)DeepSeek | 36.00 | $1122 | 1.32 | 3.96 | 103 托克/秒 |
| DeepSeek V4 Pro 0813 (Non-reasoning)DeepSeek | 20.40 | $451 | 1.32 | 3.96 | 98 托克/秒 |
| DeepSeek V4.1 Flash (Max)DeepSeek | 39.50 | $672 | 0.29 | 1.20 | 179 托克/秒 |
| DeepSeek V4.1 Flash (Non-reasoning)DeepSeek | 24.70 | $239 | 0.30 | 1.20 | 290 托克/秒 |
| DeepSeek V4.1 Flash 1017 (Max)DeepSeek | 45.10 | $445 | 0.30 | 1.20 | 186 托克/秒 |
| DeepSeek-Coder-V2DeepSeek | 6.00 | - | - | - | - |
| DeepSeek-V2-ChatDeepSeek | 5.50 | - | - | - | - |
| DeepSeek-V2.5DeepSeek | 6.60 | - | - | - | - |
| DeepSeek-V2.5 (Dec '24)DeepSeek | 6.60 | - | - | - | - |
| Devstral 2Mistral | 8.60 | - | - | - | - |
| Devstral Small (Jul '25)Mistral | 7.60 | - | - | - | - |
| Devstral Small (May '25)Mistral | 8.70 | - | - | - | - |
| Devstral Small 2Mistral | 7.50 | - | - | - | - |
| DiffusionGemma 26B A4BGoogle | 9.50 | - | - | - | - |
| ERNIE 4.5 300B A47BBaidu | 7.50 | - | - | - | - |
| Exaone 4.0 1.2B (Non-reasoning)LG AI Research | 5.20 | - | - | - | - |
| Exaone 4.0 1.2B (Reasoning)LG AI Research | 5.30 | - | - | - | - |
| EXAONE 4.0 32B (Non-reasoning)LG AI Research | 6.30 | - | - | - | - |
| EXAONE 4.0 32B (Reasoning)LG AI Research | 8.20 | - | - | - | - |
| EXAONE 4.5 33B (Non-reasoning)LG AI Research | - | - | - | - | - |
| EXAONE 4.5 33B (Reasoning)LG AI Research | 13.20 | - | - | - | - |
| Falcon-H1R-7BTII UAE | 7.80 | - | - | - | - |
| G9v3-39A5BAI9Stars | 21.80 | - | - | - | - |
| G9v3-3BAI9Stars | 10.80 | - | - | - | - |
| Gemma 3 1B InstructGoogle | 4.80 | - | 0.00 | 0.00 | - |
| Gemma 3 270MGoogle | 5.10 | - | - | - | - |
| Gemma 3 4B InstructGoogle | 4.80 | - | 0.00 | 0.00 | - |
| Gemma 3n E2B InstructGoogle | 4.80 | - | 0.00 | 0.00 | - |
| Gemma 3n E4B InstructGoogle | 4.80 | - | - | - | - |
| Gemma 3n E4B Instruct Preview (May '25)Google | 5.80 | - | - | - | - |
| Gemma 4 12B (Non-reasoning)Google | 9.40 | - | 0.10 | 0.30 | 120 托克/秒 |
| Gemma 4 12B (Reasoning)Google | 14.20 | - | 0.10 | 0.30 | 115 托克/秒 |
| Gemma 4 26B A4B (Non-reasoning)Google | 13.10 | - | 0.12 | 0.40 | 99 托克/秒 |
| Gemma 4 26B A4B (Reasoning)Google | 16.70 | - | 0.10 | 0.34 | 86 托克/秒 |
| Gemma 4 31B (Non-reasoning)Google | 13.90 | - | 0.17 | 0.40 | 53 托克/秒 |
| Gemma 4 E2B (Non-reasoning)Google | 6.50 | - | - | - | - |
| Gemma 4 E2B (Reasoning)Google | 7.80 | - | - | - | - |
| Gemma 4 E4B (Non-reasoning)Google | 7.50 | - | 0.02 | 0.10 | 49 托克/秒 |
| Gemma 4 E4B (Reasoning)Google | 8.90 | $14.7 | 0.02 | 0.10 | 44 托克/秒 |
| GLM-4.5 (Reasoning)Z AI | 12.80 | - | - | - | - |
| GLM-4.5-AirZ AI | 11.10 | - | 0.17 | 0.98 | 79 托克/秒 |
| GLM-4.5V (Non-reasoning)Z AI | 6.70 | - | 0.60 | 1.80 | 66 托克/秒 |
| GLM-4.5V (Reasoning)Z AI | 7.60 | - | 0.60 | 1.80 | 63 托克/秒 |
| GLM-4.6 (Non-reasoning)Z AI | 14.90 | - | 0.57 | 2.20 | 34 托克/秒 |
| GLM-4.6 (Reasoning)Z AI | 18.50 | - | 0.55 | 2.20 | 33 托克/秒 |
| GLM-4.6V (Non-reasoning)Z AI | 8.40 | - | 0.30 | 0.90 | 82 托克/秒 |
| GLM-4.6V (Reasoning)Z AI | 11.20 | - | 0.30 | 0.90 | 86 托克/秒 |
| GLM-4.7 (Non-reasoning)Z AI | 17.40 | - | 0.53 | 2.10 | 36 托克/秒 |
| GLM-4.7 (Reasoning)Z AI | 22.20 | - | 0.53 | 2.10 | 35 托克/秒 |
| GLM-4.7-Flash (Non-reasoning)Z AI | 10.60 | - | 0.07 | 0.40 | 99 托克/秒 |
| GLM-4.7-Flash (Reasoning)Z AI | 14.90 | - | 0.07 | 0.40 | 81 托克/秒 |
| GLM-5 (Non-reasoning)Z AI | 21.80 | - | 1.00 | 3.20 | 67 托克/秒 |
| GLM-5 (Reasoning)Z AI | 27.90 | - | 1.00 | 3.20 | 69 托克/秒 |
| GLM-5.1 (Non-reasoning)Z AI | 24.20 | - | 1.38 | 4.40 | 67 托克/秒 |
| GLM-5.1 (Reasoning)Z AI | 26.10 | $1487 | 1.28 | 4.07 | 70 托克/秒 |
| GLM-5.2 (Max)Z AI | 33.70 | $2097 | 1.40 | 4.40 | 93 托克/秒 |
| GLM-5.2 (Non-reasoning)Z AI | 22.40 | - | 1.40 | 4.40 | 181 托克/秒 |
| GLM-5.3 (Low)Z AI | 34.30 | $1013 | 1.40 | 4.40 | 89 托克/秒 |
| GLM-5.3 (Max)Z AI | 44.80 | $2503 | 1.40 | 4.40 | 86 托克/秒 |
| GLM-5.3-FlashZ AI | 41.80 | $306 | 0.15 | 0.50 | 69 托克/秒 |
| gpt-oss-120b (High)OpenAI | 11.60 | $123 | 0.15 | 0.60 | 216 托克/秒 |
| gpt-oss-120b (Low)OpenAI | 10.20 | - | 0.13 | 0.59 | 264 托克/秒 |
| gpt-oss-20b (High)OpenAI | 9.00 | $28.8 | 0.05 | 0.22 | 156 托克/秒 |
| gpt-oss-20b (Low)OpenAI | 10.00 | - | 0.07 | 0.30 | 181 托克/秒 |
| Granite 3.3 8B (Non-reasoning)IBM | 4.90 | - | 0.03 | 0.25 | 38 托克/秒 |
| Granite 4.0 1BIBM | 5.00 | - | - | - | - |
| Granite 4.0 350MIBM | 4.80 | - | - | - | - |
| Granite 4.0 H 1BIBM | 5.20 | - | - | - | - |
| Granite 4.0 H 350MIBM | 4.80 | - | - | - | - |
| Granite 4.0 H SmallIBM | 6.00 | - | - | - | - |
| Granite 4.0 MicroIBM | 5.10 | - | - | - | - |
| Granite 4.1 30BIBM | 7.40 | - | - | - | - |
| Granite 4.1 3BIBM | 5.90 | - | - | - | - |
| Granite 4.1 8BIBM | 6.60 | - | - | - | - |
| Granite 4.2 30BIBM | 12.70 | $111 | 0.16 | 0.65 | 71 托克/秒 |
| Granite 4.2 3BIBM | 9.10 | $14.9 | 0.03 | 0.12 | 72 托克/秒 |
| Granite 4.2 8BIBM | 11.10 | $53.1 | 0.06 | 0.25 | 65 托克/秒 |
| Grok 2 (Dec '24)SpaceXAI | 7.10 | - | - | - | - |
| Grok-1SpaceXAI | 6.30 | - | - | - | - |
| Hermes 3 - Llama-3.1 70BNous Research | 6.00 | - | 0.70 | 0.70 | 32 托克/秒 |
| Hermes 4 - Llama-3.1 405B (Non-reasoning)Nous Research | 7.40 | - | 1.00 | 3.00 | 43 托克/秒 |
| Hermes 4 - Llama-3.1 405B (Reasoning)Nous Research | 7.50 | - | 1.00 | 3.00 | 42 托克/秒 |
| Hermes 4 - Llama-3.1 70B (Non-reasoning)Nous Research | 6.70 | - | - | - | - |
| Hermes 4 - Llama-3.1 70B (Reasoning)Nous Research | 7.90 | - | - | - | - |
| Hy3Tencent | 25.30 | $150 | 0.14 | 0.56 | 82 托克/秒 |
| Hy3-preview (Non-reasoning)Tencent | 17.00 | - | 0.06 | 0.21 | - |
| Hy3-preview (Reasoning)Tencent | 22.70 | - | 0.06 | 0.21 | - |
| HyperCLOVA X SEED Think (32B)Naver | 11.40 | - | - | - | - |
| HyperNova 60B 2605 (High, Based on gpt-oss-120b)Multiverse Computing | 11.70 | - | - | - | - |
| Inkling (Xhigh)Thinking Machines | 25.00 | - | 1.00 | 4.05 | 198 托克/秒 |
| Inkling SmallThinking Machines | 25.70 | $241 | 0.30 | 1.20 | 226 托克/秒 |
| INTELLECT-3 (Based on GLM-4.5-Air)Prime Intellect | 10.60 | - | - | - | - |
| Jamba 1.5 LargeAI21 Labs | 6.00 | - | 2.00 | 8.00 | - |
| Jamba 1.5 MiniAI21 Labs | 5.20 | - | 0.20 | 0.40 | - |
| Jamba 1.6 LargeAI21 Labs | 6.00 | - | - | - | - |
| Jamba 1.6 MiniAI21 Labs | 5.20 | - | - | - | - |
| Jamba 1.7 LargeAI21 Labs | 6.10 | - | - | - | - |
| Jamba 1.7 MiniAI21 Labs | 5.20 | - | - | - | - |
| Jamba Reasoning 3BAI21 Labs | 5.70 | - | - | - | - |
| K-EXAONE (Non-reasoning)LG AI Research | 11.20 | - | - | - | - |
| K-EXAONE (Reasoning)LG AI Research | 14.40 | - | - | - | - |
| K-EXAONE 2.0 0803LG AI Research | 19.70 | - | - | - | - |
| K2 Horizon 0.9BFoundation Models的Institute | 3.00 | - | - | - | - |
| K2 Horizon 3.7BFoundation Models的Institute | 15.60 | - | - | - | - |
| K2 Horizon 7BFoundation Models的Institute | 20.60 | - | - | - | - |
| K2 Horizon MoVA 36B A4BFoundation Models的Institute | 25.30 | - | - | - | - |
| K2 Think V2Foundation Models的Institute | 11.50 | - | - | - | - |
| K2-V2 (High)Foundation Models的Institute | 9.90 | - | - | - | - |
| K2-V2 (Low)Foundation Models的Institute | 7.30 | - | - | - | - |
| K2-V2 (Medium)Foundation Models的Institute | 9.00 | - | - | - | - |
| Kimi K2Kimi | 12.70 | - | 0.57 | 2.30 | 48 托克/秒 |
| Kimi K2 0905Kimi | 15.30 | - | 0.60 | 2.50 | 50 托克/秒 |
| Kimi K2 ThinkingKimi | 22.00 | - | 0.60 | 2.50 | 121 托克/秒 |
| Kimi K2.5 (Non-reasoning)Kimi | 19.40 | - | 0.60 | 3.00 | 53 托克/秒 |
| Kimi K2.5 (Reasoning)Kimi | 23.50 | - | 0.60 | 3.00 | 82 托克/秒 |
| Kimi K2.6 (Non-reasoning)Kimi | 23.60 | - | 0.95 | 4.00 | 76 托克/秒 |
| Kimi K2.6 (Reasoning)Kimi | 27.00 | $1528 | 0.95 | 4.00 | 79 托克/秒 |
| Kimi K2.7 CodeKimi | 25.80 | $993 | 0.95 | 4.00 | 62 托克/秒 |
| Kimi K3 (Low)Kimi | 30.10 | $1190 | 3.00 | 15.00 | 40 托克/秒 |
| Kimi K3 (Max)Kimi | 43.60 | $3658 | 3.00 | 15.00 | 42 托克/秒 |
| Kimi Linear 48B A3B InstructKimi | 7.30 | - | - | - | - |
| LFM2 1.2BLiquid AI | 4.80 | - | - | - | - |
| LFM2 2.6BLiquid AI | 5.20 | - | - | - | - |
| LFM2 24B A2BLiquid AI | 5.90 | - | - | - | - |
| LFM2 8B A1BLiquid AI | 4.90 | - | - | - | - |
| LFM2.5-1.2B-InstructLiquid AI | 5.20 | - | - | - | - |
| LFM2.5-1.2B-ThinkingLiquid AI | 5.20 | - | - | - | - |
| LFM2.5-2.6BLiquid AI | 8.40 | - | - | - | - |
| LFM2.5-8B-A1BLiquid AI | 7.20 | - | - | - | - |
| LFM2.5-VL-1.6BLiquid AI | 4.80 | - | - | - | - |
| Ling 2.6 FlashInclusionAI | 9.70 | - | - | - | - |
| Ling 3.0 FlashInclusionAI | 20.10 | - | 0.07 | 0.22 | 320 托克/秒 |
| Ling 3.1 FlashInclusionAI | 41.10 | $1015 | 0.30 | 0.90 | 213 托克/秒 |
| Ling-1TInclusionAI | 9.20 | - | - | - | - |
| Ling-2.6-1TInclusionAI | 17.00 | - | 0.30 | 2.50 | - |
| Ling-3.0-flash-FinInclusionAI | 22.60 | - | - | - | - |
| Ling-3.0-flash-VLInclusionAI | 24.60 | - | - | - | - |
| Ling-flash-2.0InclusionAI | 7.80 | - | 0.14 | 0.57 | 5 托克/秒 |
| Ling-mini-2.0InclusionAI | 5.50 | - | - | - | - |
| Llama 2 Chat 13BMeta | 5.30 | - | - | - | - |
| Llama 2 Chat 70BMeta | 5.30 | - | - | - | - |
| Llama 2 Chat 7BMeta | 5.70 | - | 0.05 | 0.25 | - |
| Llama 3 Instruct 70BMeta | 5.50 | - | 0.65 | 2.75 | - |
| Llama 3 Instruct 8BMeta | 4.80 | - | 0.04 | 0.14 | - |
| Llama 3.1 Instruct 405BMeta | 7.30 | - | - | - | - |
| Llama 3.1 Instruct 70BMeta | 6.60 | - | 0.56 | 0.56 | 40 托克/秒 |
| Llama 3.1 Instruct 8BMeta | 6.90 | - | 0.04 | 0.07 | 135 托克/秒 |
| Llama 3.1 Nemotron Instruct 70BNVIDIA | 6.90 | - | - | - | - |
| Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)NVIDIA | 7.30 | - | - | - | - |
| Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)NVIDIA | 7.50 | - | - | - | - |
| Llama 3.1 Tulu3 405BAllen Institute for AI | 7.20 | - | - | - | - |
| Llama 3.2 Instruct 11B (Vision)Meta | 5.40 | - | 0.34 | 0.34 | 24 托克/秒 |
| Llama 3.2 Instruct 1BMeta | 4.80 | - | - | - | - |
| Llama 3.2 Instruct 3BMeta | 5.70 | - | - | - | - |
| Llama 3.2 Instruct 90B (Vision)Meta | 6.40 | - | - | - | - |
| Llama 3.3 Instruct 70BMeta | 7.70 | - | 0.66 | 0.71 | 90 托克/秒 |
| Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)NVIDIA | 7.30 | - | - | - | - |
| Llama 3.3 Nemotron Super 49B v1 (Reasoning)NVIDIA | 8.90 | - | - | - | - |
| Llama 4 MaverickMeta | 9.30 | $44.0 | 0.25 | 0.92 | 49 托克/秒 |
| Llama 4 ScoutMeta | 6.50 | $73.6 | 0.18 | 0.66 | 115 托克/秒 |
| Llama 65BMeta | 5.00 | - | - | - | - |
| Llama Nemotron Super 49B v1.5 (Non-reasoning)NVIDIA | 7.40 | - | - | - | - |
| Llama Nemotron Super 49B v1.5 (Reasoning)NVIDIA | 9.00 | - | - | - | - |
| LongCat 2.0LongCat | 19.10 | $545 | 0.75 | 2.95 | 76 托克/秒 |
| LongCat Flash LiteLongCat | 11.50 | - | - | - | - |
| Magistral Small 1Mistral | 8.20 | - | - | - | - |
| Magistral Small 1.2Mistral | 8.60 | - | 0.50 | 1.50 | 76 托克/秒 |
| MiMo-V2-Flash (Feb 2026)Xiaomi | 22.40 | - | - | - | - |
| MiMo-V2-Flash (Non-reasoning)Xiaomi | 16.00 | - | - | - | - |
| MiMo-V2-Flash (Reasoning)Xiaomi | 20.80 | - | 0.10 | 0.30 | - |
| MiMo-V2.5Xiaomi | 21.80 | $61.9 | 0.13 | 0.27 | 62 托克/秒 |
| MiMo-V2.5-Pro (Non-reasoning)Xiaomi | 18.30 | - | 0.39 | 1.17 | 52 托克/秒 |
| MiMo-V2.5-Pro (Reasoning)Xiaomi | 26.00 | $231 | 0.41 | 1.02 | 52 托克/秒 |
| MiMo-V2.6-FlashXiaomi最佳性价比前沿 | 37.90 | $110 | 0.14 | 0.28 | 56 托克/秒 |
| MiMo-V2.6-ProXiaomi最佳性价比前沿 | 46.30 | $207 | 0.43 | 0.87 | 55 托克/秒 |
| MiniCPM-V 4.6 1.3BOpenBMB | 5.70 | - | - | - | - |
| MiniCPM5-1B (Non-reasoning)OpenBMB | 8.70 | - | - | - | - |
| MiniCPM5-1B (Reasoning)OpenBMB | 8.80 | - | - | - | - |
| MiniCPM5-2BOpenBMB | 12.50 | - | - | - | - |
| MiniMax M1 40kMiniMax | 10.00 | - | - | - | - |
| MiniMax M1 80kMiniMax | 11.70 | - | 0.55 | 2.20 | 96 托克/秒 |
| MiniMax-M2MiniMax | 18.60 | - | 0.30 | 1.20 | 101 托克/秒 |
| MiniMax-M2.1MiniMax | 20.90 | - | 0.30 | 1.20 | 64 托克/秒 |
| MiniMax-M2.5MiniMax | 22.80 | - | 0.30 | 1.20 | 61 托克/秒 |
| MiniMax-M2.7MiniMax | 22.80 | - | 0.30 | 1.20 | 51 托克/秒 |
| MiniMax-M3MiniMax | 29.20 | $538 | 0.30 | 1.20 | 98 托克/秒 |
| Ministral 3 14BMistral | 6.00 | $37.6 | 0.20 | 0.20 | 83 托克/秒 |
| Ministral 3 3BMistral | 4.80 | $12.2 | 0.10 | 0.10 | 208 托克/秒 |
| Ministral 3 8BMistral | 5.50 | $27.5 | 0.15 | 0.15 | 111 托克/秒 |
| Mistral 7B InstructMistral | 5.00 | - | 0.15 | 0.20 | - |
| Mistral Large 2 (Jul '24)Mistral | 6.80 | - | 2.00 | 6.00 | - |
| Mistral Large 2 (Nov '24)Mistral | 7.60 | - | - | - | - |
| Mistral Large 3Mistral | 9.30 | $46.4 | 0.50 | 1.50 | 82 托克/秒 |
| Mistral Medium 3.5Mistral | 14.20 | $1264 | 1.50 | 7.50 | 159 托克/秒 |
| Mistral Small (Sep '24)Mistral | 5.80 | - | - | - | - |
| Mistral Small 3Mistral | 6.70 | - | 0.05 | 0.08 | 44 托克/秒 |
| Mistral Small 3.1Mistral | 7.10 | $27.8 | 0.08 | 0.14 | 45 托克/秒 |
| Mistral Small 3.2Mistral | 7.00 | $122 | 0.07 | 0.20 | 43 托克/秒 |
| Mistral Small 4 (Non-reasoning)Mistral | 9.00 | - | 0.15 | 0.60 | 174 托克/秒 |
| Mistral Small 4 (Reasoning)Mistral | 11.30 | $42.5 | 0.15 | 0.60 | 183 托克/秒 |
| Mixtral 8x22B InstructMistral | 5.70 | - | - | - | - |
| Mixtral 8x7B InstructMistral | 5.10 | - | 0.45 | 0.70 | 25 托克/秒 |
| Molmo 7B-DAllen Institute for AI | 5.60 | - | - | - | - |
| Molmo2-8BAllen Institute for AI | 5.00 | - | - | - | - |
| Motif 3Motif Technologies | 33.60 | - | - | - | - |
| Muse Glimmer (High)Meta | 17.50 | $146 | 0.30 | 1.20 | 105 托克/秒 |
| Muse Spark Open (Xhigh)Meta | 38.90 | $1185 | 1.25 | 4.25 | 151 托克/秒 |
| Nanbeige4.1-3BNanbeige | 8.40 | - | - | - | - |
| Nemotron 3 Nano Omni 30B A3B ReasoningNVIDIA | 10.30 | - | 0.40 | 1.36 | 234 托克/秒 |
| Nemotron 3 Super 120B A12B (Reasoning)NVIDIA | 12.80 | $1319 | 0.30 | 0.90 | 169 托克/秒 |
| Nemotron 3 Ultra 550B A55B (Reasoning)NVIDIA | 22.90 | $697 | 0.60 | 2.70 | 156 托克/秒 |
| Nemotron 3.5 Bolt 74B-A9BNVIDIA | 21.40 | - | - | - | - |
| Nemotron 3.5 LightningNVIDIA | 12.90 | $92.8 | 0.06 | 0.20 | 291 托克/秒 |
| Nemotron 3.5 Super VLNVIDIA | 26.50 | $286 | 0.23 | 0.57 | 444 托克/秒 |
| Nemotron Cascade 2 30B A3BNVIDIA | 11.70 | - | - | - | - |
| Nex-N2-Pro (Based on Qwen3.5-397B-A17B)Nex AGI | 28.20 | - | - | - | - |
| NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)NVIDIA | 6.80 | - | 0.05 | 0.20 | 275 托克/秒 |
| NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)NVIDIA | 8.90 | $63.7 | 0.05 | 0.20 | 235 托克/秒 |
| NVIDIA Nemotron 3 Nano 4BNVIDIA | 4.80 | - | - | - | - |
| NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)NVIDIA | 5.80 | - | 0.20 | 0.60 | 203 托克/秒 |
| NVIDIA Nemotron Nano 12B v2 VL (Reasoning)NVIDIA | 7.50 | - | - | - | - |
| NVIDIA Nemotron Nano 9B V2 (Non-reasoning)NVIDIA | 6.80 | - | 0.05 | 0.20 | 167 托克/秒 |
| NVIDIA Nemotron Nano 9B V2 (Reasoning)NVIDIA | 7.40 | - | 0.04 | 0.16 | 101 托克/秒 |
| OLMo 2 32BAllen Institute for AI | 6.00 | - | - | - | - |
| OLMo 2 7BAllen Institute for AI | 5.60 | - | - | - | - |
| Olmo 3 32B ThinkAllen Institute for AI | 6.50 | - | - | - | - |
| Olmo 3 7B InstructAllen Institute for AI | 5.20 | - | - | - | - |
| Olmo 3 7B ThinkAllen Institute for AI | 5.60 | - | - | - | - |
| Olmo 3.1 32B InstructAllen Institute for AI | 6.50 | - | - | - | - |
| Olmo 3.1 32B ThinkAllen Institute for AI | 7.10 | - | - | - | - |
| OpenChat 3.5 (1210)OpenChat | 5.30 | - | - | - | - |
| Phi-3 Mini Instruct 3.8BMicrosoft | 5.80 | - | - | - | - |
| Phi-4Microsoft | 5.90 | - | 0.13 | 0.50 | 39 托克/秒 |
| Phi-4 Mini InstructMicrosoft | 6.30 | - | 0.07 | 0.30 | 46 托克/秒 |
| Phi-4 Multimodal InstructMicrosoft | 5.80 | - | 0.08 | 0.32 | - |
| Pixtral LargeMistral | 7.10 | - | - | - | - |
| Qwen Chat 14BAlibaba | 5.00 | - | - | - | - |
| Qwen Chat 72BAlibaba | 5.40 | - | - | - | - |
| Qwen1.5 Chat 110BAlibaba | 5.70 | - | - | - | - |
| Qwen2 Instruct 72BAlibaba | 6.30 | - | - | - | - |
| Qwen2.5 Coder Instruct 32BAlibaba | 6.70 | - | - | - | - |
| Qwen2.5 Coder Instruct 7BAlibaba | 5.80 | - | - | - | - |
| Qwen2.5 Instruct 32BAlibaba | 6.90 | - | - | - | - |
| Qwen2.5 Instruct 72BAlibaba | 7.70 | - | 0.47 | 0.49 | 27 托克/秒 |
| Qwen3 0.6B (Non-reasoning)Alibaba | 4.80 | - | - | - | - |
| Qwen3 0.6B (Reasoning)Alibaba | 4.80 | - | - | - | - |
| Qwen3 1.7B (Non-reasoning)Alibaba | 4.90 | - | - | - | - |
| Qwen3 1.7B (Reasoning)Alibaba | 5.20 | - | - | - | - |
| Qwen3 14B (Non-reasoning)Alibaba | 6.70 | - | 0.12 | 0.24 | 82 托克/秒 |
| Qwen3 14B (Reasoning)Alibaba | 6.40 | - | 0.12 | 0.24 | 81 托克/秒 |
| Qwen3 235B A22B (Non-reasoning)Alibaba | 8.30 | - | - | - | - |
| Qwen3 235B A22B (Reasoning)Alibaba | 9.50 | - | - | - | - |
| Qwen3 235B A22B 2507 (Reasoning)Alibaba | 12.70 | - | 0.23 | 2.30 | 82 托克/秒 |
| Qwen3 235B A22B 2507 InstructAlibaba | 12.00 | - | 0.21 | 0.84 | 39 托克/秒 |
| Qwen3 30B A3B (Non-reasoning)Alibaba | 6.60 | - | 0.12 | 0.50 | 159 托克/秒 |
| Qwen3 30B A3B (Reasoning)Alibaba | 7.60 | - | 0.14 | 0.55 | 163 托克/秒 |
| Qwen3 30B A3B 2507 (Reasoning)Alibaba | 9.80 | - | - | - | - |
| Qwen3 30B A3B 2507 InstructAlibaba | 7.50 | - | 0.10 | 0.30 | 200 托克/秒 |
| Qwen3 32B (Non-reasoning)Alibaba | 7.30 | - | 0.15 | 0.60 | 60 托克/秒 |
| Qwen3 32B (Reasoning)Alibaba | 7.20 | - | 0.24 | 0.54 | 23 托克/秒 |
| Qwen3 4B (Non-reasoning)Alibaba | 6.60 | - | - | - | - |
| Qwen3 4B (Reasoning)Alibaba | 7.20 | - | - | - | - |
| Qwen3 4B 2507 (Reasoning)Alibaba | 8.80 | - | - | - | - |
| Qwen3 4B 2507 InstructAlibaba | 6.70 | - | - | - | - |
| Qwen3 8B (Non-reasoning)Alibaba | 6.00 | - | 0.20 | 0.20 | - |
| Qwen3 8B (Reasoning)Alibaba | 5.20 | - | - | - | - |
| Qwen3 Coder 30B A3B InstructAlibaba | 9.60 | - | 0.19 | 0.77 | 255 托克/秒 |
| Qwen3 Coder 480B A35B InstructAlibaba | 11.90 | - | 0.38 | 1.40 | 83 托克/秒 |
| Qwen3 Coder NextAlibaba | 9.20 | $405 | 0.31 | 1.00 | 91 托克/秒 |
| Qwen3 Next 80B A3B (Reasoning)Alibaba | 11.20 | - | - | - | - |
| Qwen3 Next 80B A3B InstructAlibaba | 9.60 | - | 0.10 | 1.10 | 113 托克/秒 |
| Qwen3 Omni 30B A3B (Reasoning)Alibaba | 7.80 | - | 0.25 | 0.97 | 102 托克/秒 |
| Qwen3 Omni 30B A3B InstructAlibaba | 6.00 | - | 0.25 | 0.97 | 96 托克/秒 |
| Qwen3 VL 235B A22B (Reasoning)Alibaba | 13.40 | - | - | - | - |
| Qwen3 VL 235B A22B InstructAlibaba | 9.90 | - | 0.21 | 1.39 | 41 托克/秒 |
| Qwen3 VL 30B A3B (Reasoning)Alibaba | 9.50 | - | 0.50 | 0.50 | - |
| Qwen3 VL 30B A3B InstructAlibaba | 7.90 | - | 0.35 | 0.60 | 29 托克/秒 |
| Qwen3 VL 32B (Reasoning)Alibaba | 11.90 | - | - | - | - |
| Qwen3 VL 32B InstructAlibaba | 8.40 | - | - | - | - |
| Qwen3 VL 4B (Reasoning)Alibaba | 7.00 | - | - | - | - |
| Qwen3 VL 4B InstructAlibaba | 5.70 | - | - | - | - |
| Qwen3 VL 8B (Reasoning)Alibaba | 8.20 | - | - | - | - |
| Qwen3 VL 8B InstructAlibaba | 7.30 | - | - | - | - |
| Qwen3.5 0.8B (Non-reasoning)Alibaba | 5.40 | - | - | - | - |
| Qwen3.5 0.8B (Reasoning)Alibaba | 6.10 | - | - | - | - |
| Qwen3.5 122B A10B (Non-reasoning)Alibaba | 17.70 | - | 0.40 | 3.20 | 143 托克/秒 |
| Qwen3.5 122B A10B (Reasoning)Alibaba | 15.60 | $661 | 0.40 | 3.20 | 128 托克/秒 |
| Qwen3.5 27B (Non-reasoning)Alibaba | 19.40 | - | 0.30 | 2.40 | 81 托克/秒 |
| Qwen3.5 27B (Reasoning)Alibaba | 22.90 | - | 0.30 | 2.40 | 77 托克/秒 |
| Qwen3.5 2B (Non-reasoning)Alibaba | 6.20 | - | - | - | - |
| Qwen3.5 2B (Reasoning)Alibaba | 6.90 | - | - | - | - |
| Qwen3.5 35B A3B (Non-reasoning)Alibaba | 15.10 | - | 0.25 | 2.00 | 167 托克/秒 |
| Qwen3.5 35B A3B (Reasoning)Alibaba | 19.30 | - | 0.25 | 2.00 | 145 托克/秒 |
| Qwen3.5 397B A17B (Non-reasoning)Alibaba | 21.40 | - | 0.60 | 3.60 | 59 托克/秒 |
| Qwen3.5 397B A17B (Reasoning)Alibaba | 18.40 | $941 | 0.60 | 3.60 | 59 托克/秒 |
| Qwen3.5 4B (Non-reasoning)Alibaba | 10.80 | - | 0.03 | 0.15 | 11 托克/秒 |
| Qwen3.5 4B (Reasoning)Alibaba | 13.10 | - | 0.03 | 0.15 | 15 托克/秒 |
| Qwen3.5 9B (Non-reasoning)Alibaba | 13.30 | - | 0.17 | 0.25 | 80 托克/秒 |
| Qwen3.5 9B (Reasoning)Alibaba | 11.20 | $329 | 0.14 | 0.20 | 61 托克/秒 |
| Qwen3.6 27B (Non-reasoning)Alibaba | 19.80 | - | 0.60 | 3.60 | 57 托克/秒 |
| Qwen3.6 27B (Reasoning)Alibaba | 21.40 | $1145 | 0.60 | 3.60 | 57 托克/秒 |
| Qwen3.6 35B A3B (Non-reasoning)Alibaba | 15.20 | - | 0.38 | 2.25 | 147 托克/秒 |
| Qwen3.6 35B A3B (Reasoning)Alibaba | 18.20 | $896 | 0.38 | 2.25 | 131 托克/秒 |
| Qwen3.8 2.4T A95BAlibaba | 39.90 | $2632 | 2.00 | 6.00 | 43 托克/秒 |
| Qwen3.8 27B (Low)Alibaba | 26.20 | $1060 | 0.50 | 3.00 | 50 托克/秒 |
| Qwen3.8 27B (Medium)Alibaba | 27.60 | $1165 | 0.50 | 3.00 | 49 托克/秒 |
| Qwen3.8 27B (Non-reasoning)Alibaba | 20.20 | $2022 | 0.50 | 3.00 | 49 托克/秒 |
| Qwen3.8 27B (Xhigh)Alibaba | 33.70 | $1336 | 0.50 | 3.00 | 44 托克/秒 |
| Qwen3.8-Flash-NextAlibaba | 39.80 | $363 | 0.15 | 0.47 | 52 托克/秒 |
| QwQ 32BAlibaba | 9.50 | - | - | - | - |
| QwQ 32B-PreviewAlibaba | 7.60 | - | - | - | - |
| R1 1776Perplexity | 6.40 | - | - | - | - |
| Reka Flash 3Reka AI | 5.60 | - | 0.20 | 0.80 | 88 托克/秒 |
| Ring-1TInclusionAI | 10.90 | - | - | - | - |
| Ring-2.6-1TInclusionAI | 16.60 | $576 | 0.30 | 2.50 | 121 托克/秒 |
| Ring-flash-2.0InclusionAI | 7.20 | - | - | - | - |
| Sarvam 105B (High)Sarvam | 8.80 | - | 0.04 | 0.17 | 192 托克/秒 |
| Sarvam 30B (High)Sarvam | 6.60 | - | - | - | - |
| Sarvam M (Reasoning, Based on Mistral Small 3.1)Sarvam | 5.30 | - | 0.00 | 0.00 | - |
| Seed-OSS-36B-InstructByteDance Seed | 12.10 | - | 0.21 | 0.57 | 40 托克/秒 |
| Solar MiniUpstage | 6.40 | - | 0.15 | 0.15 | - |
| Solar Open 100B (Reasoning)Upstage | 10.40 | - | - | - | - |
| Solar Open2 250BUpstage | 24.70 | - | - | - | - |
| Step 3.5 FlashStepFun | 16.60 | - | 0.10 | 0.30 | 144 托克/秒 |
| Step 3.7 FlashStepFun | 19.50 | - | 0.20 | 1.15 | 157 托克/秒 |
| Step 5 PreviewStepFun | 43.70 | $1190 | 1.00 | 2.70 | 229 托克/秒 |
| Step3 VL 10BStepFun | 7.70 | - | - | - | - |
| Tiny Aya GlobalCohere | 4.80 | - | 0.00 | 0.00 | - |
| Tri-21B-ThinkTrillion Labs | 9.00 | - | - | - | - |
| Tri-21B-think PreviewTrillion Labs | 9.60 | - | - | - | - |
| Trinity Large ThinkingArcee AI | 10.80 | $233 | 0.25 | 0.80 | 355 托克/秒 |
使用指南
我们如何比较模型
智能得分来自 Artificial Analysis Intelligence Index。基准测试费用是运行该测试的美元总费用，包含推理 token，并非 API 的单个 token 价格。所有语言版本使用相同的数据和计算方法。综合得分不能代表模型在特定语言或任务上的表现。