LLM 排行榜——比较 OpenAI、Anthropic、Google、SpaceXAI 等公司的 AI 模型
比较并排名 250 多个 AI 模型（LLM）在智能、价格、性能与速度（输出速度——每秒 token 数；延迟——TTFT）、上下文窗口等关键指标上的表现。
如需了解包括方法论在内的更多详情，请参阅我们的常见问题。
智能
Claude Opus 5.5 (max with fallback) 和 Claude Sonnet 5.5 (max with fallback) 是智能得分最高的模型，之后是 Claude Opus 5.5 (xhigh with fallback) 和 Muse Wonder Full (max)。
输出速度
Celeris-1 和 Mercury 2.5 是速度最快的模型，之后是 Gemini 3.5 Flash-Lite 和 Trinity Large Thinking。
延迟
North Mini Code 和 Nemotron 3 Nano Omni 30B A3B 是延迟最低的模型，之后是 Command A+ 和 NVIDIA Nemotron Nano 9B V2。
每任务成本
K2 Horizon 375B A23B 和 Gemma 4 31B 的每任务成本最低，之后是 Command A+ 和 Ling 3.0 Tiny。
上下文窗口
Llama 4 Scout 和 GPT-6 Sol (Daybreak Blue, max) 支持最大的上下文窗口，之后是 Muse Wonder (max) 和 Kimi K3 (max)。
进一步分析 | ||||||||
|---|---|---|---|---|---|---|---|---|
Claude Opus 5.5 (max with fallback) | 1M | Anthropic | 58 | $5.98 | 96 | 716.75 | 721.98 | |
Claude Sonnet 5.5 (max with fallback) | 1M | Anthropic | 56 | $5.46 | 135 | 453.43 | 457.12 | |
Claude Opus 5.5 (xhigh with fallback) | 1M | Anthropic | 56 | $3.46 | 81 | 159.41 | 165.54 | |
Muse Wonder Full (max) | 1M | Meta | 56 | -- | -- | -- | -- | |
Claude Opus 5.5 (high with fallback) | 1M | Anthropic | 54 | $1.82 | 82 | 52.59 | 58.67 | |
Claude Fable 5.1 (max with fallback) | 1M | Anthropic | 53 | $7.63 | 67 | 246.61 | 254.06 | |
Claude Fable 5.1 (xhigh with fallback) | 1M | Anthropic | 53 | $5.98 | 62 | 113.74 | 121.85 | |
GPT-6 Astra (max) | 1M | OpenAI | 53 | $3.26 | 66 | 359.94 | 367.53 | |
Gemini 4 Argon (high) | 1M | Google | 53 | $1.99 | -- | -- | -- | |
Muse Wonder (max) | 1.05M | Meta | 52 | $4.52 | 88 | 19.27 | 47.72 | |
GPT-6 Astra (xhigh) | 1M | OpenAI | 52 | $2.31 | 60 | 169.96 | 178.35 | |
Claude Sonnet 5.5 (xhigh with fallback) | 1M | Anthropic | 52 | $2.01 | 103 | 37.67 | 42.51 | |
GPT-6.1 Sol (max) | 1M | OpenAI | 52 | $0.72 | 65 | 330.58 | 338.25 | |
Claude Opus 5.5 (medium with fallback) | 1M | Anthropic | 51 | $1.34 | 79 | 25.47 | 31.82 | |
Claude Fable 5.1 (high with fallback) | 1M | Anthropic | 51 | $3.91 | 52 | 17.98 | 27.56 | |
GPT-6.1 Sol (xhigh) | 1M | OpenAI | 51 | $0.39 | 59 | 177.62 | 186.14 | |
GPT-6 Astra (high) | 1M | OpenAI | 51 | $1.73 | 55 | 111.28 | 120.38 | |
GPT-6.1 Sol (high) | 1M | OpenAI | 50 | $0.32 | 57 | 59.66 | 68.44 | |
GPT-6 Astra (medium) | 1M | OpenAI | 50 | $1.54 | 55 | 6.27 | 15.42 | |
Claude Fable 5.1 (medium with fallback) | 1M | Anthropic | 49 | $2.98 | 54 | 12.09 | 21.32 | |
Muse Spark 1.3 (max) | 1M | Meta | 48 | $1.60 | 124 | 40.76 | 60.93 | |
GPT-6.1 Sol (medium) | 1M | OpenAI | 48 | $0.21 | 55 | 7.64 | 16.70 | |
Claude Fable 5.1 (low with fallback) | 1M | Anthropic | 47 | $2.37 | 56 | 2.04 | 11.02 | |
Claude Sonnet 5.5 (high with fallback) | 1M | Anthropic | 47 | $0.88 | 96 | 13.91 | 19.12 | |
Grok 4.7 (xhigh) | 500k |  SpaceXAI | 46 | $3.74 | 84 | 88.77 | 94.74 | |
Grok 4.7 (high) | 500k | SpaceXAI | 46 | $2.73 | 74 | 46.44 | 53.22 | |
MiMo-V2.6-Pro | 1M |  Xiaomi | 46 | $0.13 | 48 | 3.74 | 55.92 | |
GPT-6 Astra (low) | 1M | OpenAI | 46 | $0.82 | 54 | 2.47 | 11.77 | |
Qwen3.8 Max (0902) | 984k |  Alibaba | 45 | $5.41 | 41 | 2.89 | 64.54 | |
Muse Spark 1.3 (xhigh) | 1M | Meta | 45 | $1.37 | 245 | 21.50 | 31.72 | |
GLM-5.3 (max) | 1M |  Z AI | 45 | $2.01 | 66 | 3.49 | 41.62 | |
Step 5 Preview | 1M |  StepFun | 44 | $1.03 | 89 | 2.93 | 30.95 | |
Kimi K3 (max) | 1.05M | Kimi | 44 | $2.00 | 42 | 3.18 | 62.53 | |
Claude Haiku 5.5 (max) | 1M | Anthropic | 43 | $0.21 | 236 | 309.80 | 311.92 | |
Claude Opus 5.5 (low with fallback) | 1M | Anthropic | 42 | $0.55 | 80 | 12.35 | 18.60 | |
Grok 4.7 (low) | 500k | SpaceXAI | 42 | $1.25 | 65 | 6.97 | 14.68 | |
GPT-6.1 Sol (low) | 1M | OpenAI | 42 | $0.13 | 52 | 2.40 | 11.98 | |
GPT-5.6 Terra (max) | 1M | OpenAI | 42 | $1.40 | 126 | 129.41 | 133.39 | |
GLM-5.3-Flash | 1M | Z AI | 42 | $0.28 | 62 | 2.89 | 43.24 | |
Claude Haiku 5.5 (xhigh) | 1M | Anthropic | 41 | $0.12 | 187 | 79.11 | 81.78 | |
Ling 3.1 Flash | 1M | InclusionAI | 41 | $1.00 | 212 | 1.87 | 13.67 | |
Gemini 3.8 Flash (high) | 1M | Google | 41 | $1.24 | 137 | 21.42 | 25.08 | |
Claude Sonnet 5.5 (medium with fallback) | 1M | Anthropic | 41 | $0.48 | 95 | 0.87 | 6.14 | |
Qwen3.8 2.4T A95B | 262k | Alibaba | 40 | $2.16 | 39 | 3.10 | 66.63 | |
Qwen3.8-Flash-Next | 256k | Alibaba | 40 | $0.37 | 51 | 2.02 | 51.41 | |
Gemini 3.8 Flash (medium) | 1M | Google | 40 | $0.93 | -- | -- | -- | |
DeepSeek V4.1 Flash (max) | 1M |  DeepSeek | 39 | $0.27 | 199 | 1.02 | 13.57 | |
Mistral Large 4 Preview | 524k |  Mistral | 38 | $1.13 | -- | -- | -- | |
GPT-6 Luna (max) | 1M | OpenAI | 38 | $0.07 | 134 | 120.25 | 123.97 | |
GPT-5.6 Terra (xhigh) | 1M | OpenAI | 38 | $0.63 | 110 | 43.03 | 47.56 | |
MiMo-V2.6-Flash | 1M | Xiaomi | 38 | $0.06 | 49 | 3.21 | 54.01 | |
Claude Haiku 5.5 (high) | 1M | Anthropic | 38 | $0.08 | 175 | 24.59 | 27.45 | |
DeepSeek V4 Pro 0813 (max) | 1M | DeepSeek | 36 | $0.67 | 92 | 1.64 | 28.90 | |
Claude Sonnet 5.5 (low with fallback) | 1M | Anthropic | 36 | $0.35 | 100 | 0.57 | 5.56 | |
DeepSeek V4 Flash Vision (max) | 1M | DeepSeek | 35 | $0.31 | 189 | 1.03 | 14.26 | |
GPT-6 Luna (xhigh) | 1M | OpenAI | 35 | $0.04 | 126 | 41.65 | 45.61 | |
Claude Haiku 5.5 (medium) | 1M | Anthropic | 34 | $0.05 | 171 | 12.73 | 15.65 | |
GLM-5.3 (low) | 1M | Z AI | 34 | $0.85 | 73 | 2.99 | 37.38 | |
GPT-5.6 Terra (high) | 1M | OpenAI | 34 | $0.34 | 102 | 7.36 | 12.27 | |
JT-4.1 Flash 236B A21B | 262k |  China Mobile | 34 | -- | -- | -- | -- | |
Qwen3.8 27B (xhigh) | 256k | Alibaba | 34 | $1.01 | 46 | 3.79 | 58.28 | |
Motif 3 | 262k |  Motif Technologies | 34* | -- | -- | -- | -- | |
Gemini 3.8 Flash (low) | 1M | Google | 33 | -- | -- | -- | -- | |
GPT-6 Luna (high) | 1M | OpenAI | 33 | $0.03 | 134 | 19.68 | 23.42 | |
GPT-5.3 Codex (xhigh) | 400k | OpenAI | 33* | -- | 113 | 70.37 | 74.79 | |
K2 Horizon 375B A23B | 524k | Institute of Foundation Models | 31 | $0.00 | 119 | 3.98 | 25.05 | |
GPT-5.6 Terra (medium) | 1M | OpenAI | 30 | $0.18 | 103 | 1.75 | 6.63 | |
Kimi K3 (low) | 1.05M | Kimi | 30 | $1.15 | 45 | 3.24 | 58.97 | |
GPT-6 Luna (medium) | 1M | OpenAI | 30 | $0.02 | -- | -- | -- | |
Gemini 3.1 Pro Preview | 1M | Google | 30 | $1.30 | 134 | 20.91 | 24.64 | |
Claude Haiku 5.5 (low) | 1M | Anthropic | 29 | $0.02 | 164 | 8.80 | 11.85 | |
MiniMax-M3 | 1M |  MiniMax | 29 | $0.51 | 107 | 0.89 | 24.21 | |
Nex-N2-Pro | 262k |  Nex AGI | 28* | -- | -- | -- | -- | |
Solar Pro 4 | 512k |  Upstage | 28* | -- | 95 | 2.08 | 28.29 | |
Qwen3.8 27B (medium) | 256k | Alibaba | 28 | $1.13 | 50 | 3.92 | 54.31 | |
GPT-5.6 Terra (low) | 1M | OpenAI | 27 | $0.14 | 100 | 1.48 | 6.48 | |
JT-4.1 Flash 236B A21B (non-reasoning) | 262k | China Mobile | 27* | -- | -- | -- | -- | |
Quasar 438B (max) | 1M |  Multiverse Computing | 27 | $2.02 | 164 | 1.26 | 16.48 | |
Apodex 1.1 | 256k |  Apodex | 26 | $0.46 | -- | -- | -- | |
Qwen3.8 27B (low) | 256k | Alibaba | 26 | $1.05 | 53 | 3.78 | 51.22 | |
GPT-5.5 Instant (June 2026) | 400k | OpenAI | 26 | $0.69 | 152 | 0.90 | 17.36 | |
Kimi K2.7 Code | 256k | Kimi | 26 | $0.54 | 75 | 2.80 | 39.28 | |
Inkling Small | 1M |  Thinking Machines | 26 | $0.09 | 200 | 2.20 | 14.70 | |
K2 Horizon MoVA 36B A4B | 524k | Institute of Foundation Models | 25 | -- | -- | -- | -- | |
Hy3 | 256k |  Tencent | 25 | $0.07 | 84 | 2.96 | 32.72 | |
Qwen3.7 Plus | 1M | Alibaba | 25 | $0.22 | 55 | 2.21 | 47.51 | |
Inkling (xhigh) | 1M | Thinking Machines | 25 | -- | 186 | 1.81 | 15.28 | |
Solar Open2 250B | 1.05M | Upstage | 25* | -- | -- | -- | -- | |
DeepSeek V4.1 Flash (non-reasoning) | 1M | DeepSeek | 25 | $0.15 | 193 | 1.12 | 3.70 | |
Ling-3.0-flash-VL | 262k | InclusionAI | 25 | -- | -- | -- | -- | |
Solar Mini 4 | 1.05M | Upstage | 24 | $0.37 | 78 | 1.82 | 33.76 | |
Nemotron 3 Ultra | 262k | NVIDIA | 23 | $0.60 | 159 | 1.09 | 18.53 | |
Ling-3.0-flash-Fin | 262k | InclusionAI | 23 | -- | 317 | 1.68 | 9.56 | |
Gemini 3.5 Flash-Lite | 1M | Google | 22 | $0.19 | 408 | 7.36 | 8.58 | |
G9v3-39A5B | 131k |  AI9Stars | 22* | -- | -- | -- | -- | |
KAT-Coder-Pro V2 | 256k |  KwaiKAT | 22* | -- | -- | -- | -- | |
GPT-6 Luna (low) | 1M | OpenAI | 22 | $0.0045 | 126 | 2.55 | 6.52 | |
A.X-K2 | 262k |  SK Telecom | 21 | -- | -- | -- | -- | |
Qwen3.5 397B A17B (non-reasoning) | 262k | Alibaba | 21* | -- | 84 | 2.29 | 8.25 | |
GPT-5.6 Terra (non-reasoning) | 1M | OpenAI | 21 | $0.14 | 100 | 0.69 | 5.67 | |
K2 Horizon 7B | 524k | Institute of Foundation Models | 21 | -- | -- | -- | -- | |
Qwen3.5 Omni Plus | 256k | Alibaba | 20* | -- | 88 | 2.08 | 7.73 | |
DeepSeek V4 Pro 0813 (non-reasoning) | 1M | DeepSeek | 20 | $0.36 | 92 | 1.64 | 7.05 | |
o3 | 200k | OpenAI | 20* | -- | 129 | 6.63 | 10.51 | |
Qwen3.8 27B (non-reasoning) | 256k | Alibaba | 20 | $2.49 | 52 | 3.91 | 13.60 | |
K-EXAONE 2.0 | 262k |  LG AI Research | 20* | -- | -- | -- | -- | |
LongCat 2.0 | 1M |  LongCat | 19 | $0.15 | 65 | 2.68 | 40.92 | |
JT-35B-Flash | 256k | China Mobile | 19* | -- | -- | -- | -- | |
GPT-6 Luna (non-reasoning) | 1M | OpenAI | 18 | $0.01 | 114 | 0.83 | 5.22 | |
Qwen3.5 397B A17B | 262k | Alibaba | 18 | $0.47 | 87 | 2.31 | 44.69 | |
Qwen3.6 35B A3B | 262k | Alibaba | 18 | $0.48 | 129 | 2.05 | 47.62 | |
Qwen3.5 122B A10B (non-reasoning) | 262k | Alibaba | 18* | -- | 143 | 2.30 | 5.80 | |
Muse Glimmer (high) | 131k | Meta | 17 | $0.06 | 134 | 0.99 | 19.68 | |
Doubao Seed Code | 256k |  ByteDance Seed | 17* | -- | -- | -- | -- | |
Gemma 4 26B A4B | 256k | Google | 17* | -- | 75 | 0.97 | 34.49 | |
Ring-2.6-1T | 262k | InclusionAI | 17 | $0.29 | 119 | 3.68 | 24.61 | |
K2 Horizon 3.7B | 524k | Institute of Foundation Models | 16 | -- | -- | -- | -- | |
Qwen3.5 122B A10B | 262k | Alibaba | 16 | $0.32 | 128 | 2.35 | 21.85 | |
Qwen3.6 35B A3B (non-reasoning) | 262k | Alibaba | 15* | -- | 137 | 2.01 | 5.65 | |
Gemma 4 31B | 256k | Google | 15 | $0.00 | 35 | 1.08 | 64.72 | |
ERNIE 5.0 Thinking Preview | 128k |  Baidu | 14* | -- | -- | -- | -- | |
Mistral Medium 3.5 | 256k | Mistral | 14 | $0.50 | 163 | 2.17 | 17.48 | |
Gemma 4 12B | 262k | Google | 14* | -- | 111 | 2.51 | 25.06 | |
Nova 2.0 Pro Preview (medium) | 256k |  Amazon | 14* | -- | 122 | 13.15 | 33.57 | |
Gemma 4 31B (non-reasoning) | 256k | Google | 14* | -- | 46 | 1.96 | 12.83 | |
Nova 2.0 Omni (medium) | 1M | Amazon | 14* | -- | -- | -- | -- | |
Apriel-v1.6-15B-Thinker | 128k |  ServiceNow | 13* | -- | -- | -- | -- | |
Nova 2.0 Lite (high) | 1M | Amazon | 13* | -- | 188 | 19.63 | 32.96 | |
Qwen3.5 9B (non-reasoning) | 262k | Alibaba | 13* | -- | 98 | 0.72 | 5.84 | |
EXAONE 4.5 33B | 262k | LG AI Research | 13* | -- | -- | -- | -- | |
Command A+ | 192k | Cohere | 13 | $0.00 | 175 | 0.50 | 14.81 | |
Gemma 4 26B A4B (non-reasoning) | 256k | Google | 13* | -- | 98 | 0.99 | 6.10 | |
Qwen3.5 4B | 262k | Alibaba | 13* | -- | 23 | 0.81 | 107.53 | |
Nemotron 3.5 Lightning | 1M | NVIDIA | 13 | $0.09 | 302 | 0.62 | 8.90 | |
Nemotron 3 Super | 1M | NVIDIA | 13 | $1.64 | 152 | 1.30 | 17.72 | |
Nova 2.0 Pro Preview (low) | 256k | Amazon | 13* | -- | 120 | 9.70 | 30.53 | |
Granite 4.2 30B | 131k |  IBM | 13 | $0.07 | 77 | 0.89 | 33.47 | |
Nova 2.0 Lite (medium) | 1M | Amazon | 12* | -- | 188 | 18.07 | 31.34 | |
Qwen3.5 Omni Flash | 256k | Alibaba | 12* | -- | 226 | 1.79 | 3.99 | |
MiniCPM5-2B | 131k |  OpenBMB | 12 | -- | -- | -- | -- | |
Mercury 2.5 | 260k | Inception | 12 | $0.12 | 689 | 3.38 | 4.11 | |
JT-MINI | 128k | China Mobile | 12* | -- | -- | -- | -- | |
Magistral Medium 1.2 | 128k | Mistral | 12* | -- | -- | -- | -- | |
Nova 2.0 Lite (low) | 1M | Amazon | 12* | -- | 178 | 10.37 | 24.37 | |
HyperNova 60B 2605 (high) | 131k | Multiverse Computing | 12* | -- | -- | -- | -- | |
Nemotron Cascade 2 30B A3B | 1M | NVIDIA | 12* | -- | -- | -- | -- | |
gpt-oss-120b (high) | 131k | OpenAI | 12 | $0.11 | 242 | 0.82 | 11.16 | |
K2 Think V2 | 262k | Institute of Foundation Models | 11* | -- | -- | -- | -- | |
LongCat Flash Lite | 256k | LongCat | 11* | -- | -- | -- | -- | |
HyperCLOVA X SEED Think (32B) | 128k |  Naver | 11* | -- | -- | -- | -- | |
Mistral Small 4 | 256k | Mistral | 11 | $0.02 | 158 | 0.79 | 16.63 | |
Qwen3 Next 80B A3B | 262k | Alibaba | 11* | -- | 159 | 0.70 | 16.42 | |
Qwen3.5 9B | 262k | Alibaba | 11 | $0.21 | 58 | 1.73 | 44.64 | |
Nova 2.0 Omni (low) | 1M | Amazon | 11* | -- | -- | -- | -- | |
Granite 4.2 8B | 131k | IBM | 11 | $0.02 | 59 | 0.75 | 43.17 | |
Ling 3.0 Tiny | 262k | InclusionAI | 11 | $0.00 | 18 | 5.92 | 144.12 | |
Mi:dm K 2.5 Pro | 128k |  Korea Telecom | 11* | -- | -- | -- | -- | |
G9v3-3B | 131k | AI9Stars | 11* | -- | -- | -- | -- | |
Trinity Large Thinking | 512k | Arcee AI | 11 | $0.12 | 346 | 1.01 | 8.24 | |
Qwen3.5 4B (non-reasoning) | 262k | Alibaba | 11* | -- | 22 | 0.77 | 23.82 | |
INTELLECT-3 | 131k |  Prime Intellect | 11* | -- | -- | -- | -- | |
Solar Open 100B | 128k | Upstage | 10* | -- | -- | -- | -- | |
Nemotron 3 Nano Omni 30B A3B | 256k | NVIDIA | 10* | -- | 243 | 0.42 | 10.71 | |
gpt-oss-120b (low) | 131k | OpenAI | 10* | -- | 244 | 0.82 | 11.08 | |
Llama 4 Maverick | 1M | Meta | 10* | -- | 45 | 1.09 | 12.08 | |
Nova 2.0 Pro Preview (non-reasoning) | 256k | Amazon | 10* | -- | 111 | 1.04 | 5.53 | |
gpt-oss-20b (low) | 131k | OpenAI | 10* | -- | 296 | 0.76 | 9.21 | |
North Mini Code | 256k | Cohere | 10 | $0.00 | 119 | 0.40 | 21.45 | |
K2-V2 (high) | 512k | Institute of Foundation Models | 10* | -- | -- | -- | -- | |
Qwen3 Next 80B A3B | 262k | Alibaba | 10* | -- | 161 | 0.83 | 3.93 | |
DiffusionGemma 26B A4B | 256k | Google | 10* | -- | -- | -- | -- | |
Gemma 4 12B (non-reasoning) | 262k | Google | 9* | -- | 114 | 2.52 | 6.92 | |
Qwen3 Coder Next | 256k | Alibaba | 9 | $0.78 | 73 | 1.07 | 7.95 | |
Motif-2-12.7B | 128k | Motif Technologies | 9* | -- | -- | -- | -- | |
Nova Premier | 1M | Amazon | 9* | -- | -- | -- | -- | |
Granite 4.2 3B | 131k | IBM | 9 | $0.01 | 214 | 0.66 | 12.36 | |
K2-V2 (medium) | 512k | Institute of Foundation Models | 9* | -- | -- | -- | -- | |
Mistral Small 4 (non-reasoning) | 256k | Mistral | 9* | -- | 153 | 0.88 | 4.15 | |
Tri-21B-Think | 32k |  Trillion Labs | 9* | -- | -- | -- | -- | |
gpt-oss-20b (high) | 131k | OpenAI | 9 | $0.01 | 223 | 0.74 | 11.93 | |
Gemma 4 E4B | 128k | Google | 9* | -- | 35 | 0.95 | 72.55 | |
Nemotron 3 Nano | 1M | NVIDIA | 9 | $0.02 | 228 | 0.97 | 11.94 | |
MiniCPM5-1B | 128k | OpenBMB | 9* | -- | -- | -- | -- | |
Sarvam 105B (high) | 128k |  Sarvam | 9* | -- | -- | -- | -- | |
Nova 2.0 Lite (non-reasoning) | 1M | Amazon | 9* | -- | 174 | 1.06 | 3.93 | |
MiniCPM5-1B (non-reasoning) | 128k | OpenBMB | 9* | -- | -- | -- | -- | |
Magistral Small 1.2 | 128k | Mistral | 9* | -- | 75 | 1.59 | 34.77 | |
Nanbeige4.1-3B | 256k |  Nanbeige | 8* | -- | -- | -- | -- | |
LFM2.5-2.6B | 128k |  Liquid AI | 8* | -- | -- | -- | -- | |
EXAONE 4.0 32B | 131k | LG AI Research | 8* | -- | -- | -- | -- | |
Nova 2.0 Omni (non-reasoning) | 1M | Amazon | 8* | -- | -- | -- | -- | |
Llama 4 Scout | 10M | Meta | 8* | -- | 67 | 0.88 | 8.32 | |
Hermes 4 70B | 128k |  Nous Research | 8* | -- | -- | -- | -- | |
Falcon-H1R-7B | 256k |  TII UAE | 8* | -- | -- | -- | -- | |
Gemma 4 E2B | 128k | Google | 8* | -- | -- | -- | -- | |
Qwen3 Omni 30B A3B | 65.5k | Alibaba | 8* | -- | 103 | 1.93 | 26.30 | |
Step3 VL 10B | 65.5k | StepFun | 8* | -- | -- | -- | -- | |
Llama 3.3 70B | 128k | Meta | 8* | -- | 92 | 1.70 | 7.13 | |
Llama Nemotron Ultra | 128k | NVIDIA | 8* | -- | -- | -- | -- | |
ERNIE 4.5 300B A47B | 131k | Baidu | 8* | -- | -- | -- | -- | |
Hermes 4 405B | 128k | Nous Research | 7* | -- | 41 | 2.35 | 63.79 | |
NVIDIA Nemotron Nano 12B v2 VL | 128k | NVIDIA | 7* | -- | -- | -- | -- | |
Gemma 4 E4B (non-reasoning) | 128k | Google | 7* | -- | 31 | 0.98 | 16.97 | |
NVIDIA Nemotron Nano 9B V2 | 131k | NVIDIA | 7* | -- | 202 | 0.53 | 12.90 | |
Hermes 4 405B (non-reasoning) | 128k | Nous Research | 7* | -- | 38 | 2.34 | 15.40 | |
K2-V2 (low) | 512k | Institute of Foundation Models | 7* | -- | -- | -- | -- | |
Kimi Linear 48B A3B Instruct | 1M | Kimi | 7* | -- | -- | -- | -- | |
Llama 3.1 405B | 128k | Meta | 7* | -- | -- | -- | -- | |
LFM2.5-8B-A1B | 32.8k | Liquid AI | 7* | -- | -- | -- | -- | |
Ring-flash-2.0 | 128k | InclusionAI | 7* | -- | -- | -- | -- | |
Olmo 3.1 32B Think | 65.5k |  Allen Institute for AI | 7* | -- | -- | -- | -- | |
Command A | 256k | Cohere | 7* | -- | 60 | 1.81 | 10.11 | |
Qwen3.5 2B | 262k | Alibaba | 7* | -- | -- | -- | -- | |
Nemotron 3 Nano (non-reasoning) | 1M | NVIDIA | 7* | -- | 248 | 0.56 | 2.58 | |
NVIDIA Nemotron Nano 9B V2 (non-reasoning) | 131k | NVIDIA | 7* | -- | 167 | 1.08 | 4.06 | |
Hermes 4 70B (non-reasoning) | 128k | Nous Research | 7* | -- | -- | -- | -- | |
Sarvam 30B (high) | 65.5k | Sarvam | 7* | -- | -- | -- | -- | |
Olmo 3.1 32B Instruct | 65.5k | Allen Institute for AI | 6* | -- | -- | -- | -- | |
Gemma 4 E2B (non-reasoning) | 128k | Google | 6* | -- | -- | -- | -- | |
R1 1776 | 128k |  Perplexity | 6* | -- | -- | -- | -- | |
Llama 3.2 90B (Vision) | 128k | Meta | 6* | -- | -- | -- | -- | |
Celeris-1 | 131k | Celeris | 6 | $0.05 | 1,522 | 0.58 | 0.91 | |
Phi-4 Mini | 128k |  Microsoft | 6* | -- | 46 | 0.82 | 11.67 | |
EXAONE 4.0 32B (non-reasoning) | 131k | LG AI Research | 6* | -- | -- | -- | -- | |
Qwen3.5 2B (non-reasoning) | 262k | Alibaba | 6* | -- | -- | -- | -- | |
Qwen3.5 0.8B | 262k | Alibaba | 6* | -- | -- | -- | -- | |
DeepHermes 3 - Mistral 24B (non-reasoning) | 32k | Nous Research | 6* | -- | -- | -- | -- | |
Jamba 1.7 Large | 256k |  AI21 Labs | 6* | -- | -- | -- | -- | |
Granite 4.0 H Small | 128k | IBM | 6* | -- | 14 | 35.61 | 70.72 | |
Ministral 3 14B | 256k | Mistral | 6 | $0.02 | 57 | 1.09 | 9.84 | |
Qwen3 Omni 30B A3B | 65.5k | Alibaba | 6* | -- | 96 | 1.86 | 7.09 | |
LFM2 24B A2B | 32.8k | Liquid AI | 6* | -- | -- | -- | -- | |
Phi-4 | 16k | Microsoft | 6* | -- | 42 | 2.45 | 14.41 | |
Nova Micro | 130k | Amazon | 6* | -- | 282 | 0.87 | 2.64 | |
NVIDIA Nemotron Nano 12B v2 VL (non-reasoning) | 128k | NVIDIA | 6* | -- | 207 | 1.15 | 3.57 | |
Phi-4 Multimodal | 128k | Microsoft | 6* | -- | -- | -- | -- | |
MiniCPM-V 4.6 1.3B | 262k | OpenBMB | 6* | -- | -- | -- | -- | |
Jamba Reasoning 3B | 262k | AI21 Labs | 6* | -- | -- | -- | -- | |
Reka Flash 3 | 128k |  Reka AI | 6* | -- | 93 | 1.40 | 28.31 | |
Olmo 3 7B Think | 65.5k | Allen Institute for AI | 6* | -- | -- | -- | -- | |
Molmo 7B-D | 4.1k | Allen Institute for AI | 6* | -- | -- | -- | -- | |
Ministral 3 8B | 256k | Mistral | 5 | $0.01 | 107 | 0.73 | 5.42 | |
Llama 3.2 11B (Vision) | 128k | Meta | 5* | -- | 22 | 1.73 | 24.76 | |
Qwen3.5 0.8B (non-reasoning) | 262k | Alibaba | 5* | -- | -- | -- | -- | |
Exaone 4.0 1.2B | 64k | LG AI Research | 5* | -- | -- | -- | -- | |
Olmo 3 7B | 65.5k | Allen Institute for AI | 5* | -- | -- | -- | -- | |
Exaone 4.0 1.2B (non-reasoning) | 64k | LG AI Research | 5* | -- | -- | -- | -- | |
LFM2.5-1.2B-Thinking | 32k | Liquid AI | 5* | -- | -- | -- | -- | |
Jamba 1.7 Mini | 258k | AI21 Labs | 5* | -- | -- | -- | -- | |
LFM2.5-1.2B-Instruct | 32k | Liquid AI | 5* | -- | -- | -- | -- | |
Granite 4.0 H 1B | 128k | IBM | 5* | -- | -- | -- | -- | |
Gemma 3 270M | 32k | Google | 5* | -- | -- | -- | -- | |
Apertus 70B Instruct | 65.5k |  Swiss AI Initiative | 5* | -- | -- | -- | -- | |
Granite 4.0 Micro | 128k | IBM | 5* | -- | -- | -- | -- | |
DeepHermes 3 - Llama-3.1 8B (non-reasoning) | 128k | Nous Research | 5* | -- | -- | -- | -- | |
Molmo2-8B | 36.9k | Allen Institute for AI | 5* | -- | -- | -- | -- | |
Ministral 3 3B | 256k | Mistral | 5 | $0.01 | 215 | 0.62 | 2.95 | |
LFM2.5-VL-1.6B | 32k | Liquid AI | 5* | -- | -- | -- | -- | |
Granite 4.0 350M | 32.8k | IBM | 5* | -- | -- | -- | -- | |
Tiny Aya Global | 8.19k | Cohere | 5* | -- | -- | -- | -- | |
Apertus 8B Instruct | 65.5k | Swiss AI Initiative | 5* | -- | -- | -- | -- | |
Granite 4.0 H 350M | 32.8k | IBM | 5* | -- | -- | -- | -- | |
Nemotron 3 Nano 4B | 262k | NVIDIA | 5 | -- | -- | -- | -- | |
K2 Horizon 0.9B | 131k | Institute of Foundation Models | 3 | -- | -- | -- | -- | |
EXAONE 4.5 33B (non-reasoning) | 262k | LG AI Research | -- | -- | -- | -- | -- | |
GPT-6 Sol (Daybreak Blue, max) | 1.05M | OpenAI | -- | -- | -- | -- | -- | |
Gemini 3 Deep Think | 128k | Google | -- | -- | -- | -- | -- | |
GPT-5.5 Pro (xhigh) | 922k | OpenAI | -- | -- | -- | -- | -- | |
Cogito v2.1 | 128k |  Deep Cogito | -- | -- | -- | -- | -- | |
关键定义
常见问题
在参与排名的 127 个模型中，Claude Opus 5.5 (Max, Default Fallback) 目前以 58 的 Intelligence Index 得分位居 Artificial Analysis LLM 排行榜第一。
按 Intelligence Index 排名，顶尖模型是：1. Claude Opus 5.5 (Max, Default Fallback)（58）, 2. Claude Sonnet 5.5 (Max, Default Fallback)（56）, 3. Claude Opus 5.5 (Xhigh, Default Fallback)（56）, 4. Muse Wonder Full (Max)（56）, 5. Claude Opus 5.5 (High, Default Fallback)（54）。
Celeris-1 最快，输出速度为每秒 1,522.5 个 token，其次是 Mercury 2.5（688.9 t/s）和 Gemini 3.5 Flash-Lite（407.6 t/s）。
K2 Horizon 375B A23B 的每项 Intelligence Index 任务成本最低，为 $0.00，其次是 Gemma 4 31B (Reasoning)（$0.00）和 Command A+（$0.00）。
MiMo-V2.6-Pro 是排名最高的开放权重模型，Intelligence Index 得分为 46。排行榜上的 127 个模型中，有 59 个开放权重模型。
按 Intelligence Index 排名，顶尖开放权重模型是：1. MiMo-V2.6-Pro（46）, 2. GLM-5.3 (Max)（45）, 3. Kimi K3 (Max)（44）。
在 117 个推理模型中，Claude Opus 5.5 (Max, Default Fallback) 以 58 的 Intelligence Index 得分领先。推理模型会先通过扩展思考来解决复杂问题，再给出回答。
排行榜提供筛选条件，可按模型类型（推理与非推理）、开放性（开放权重与专有）及其他标准缩小结果范围。您还可以调整提示词选项，查看不同输入长度下的性能变化。
点击排行榜中的任意模型名称，即可访问其专属比较页面，查看智能、价格、速度、延迟等详细图表。您还可以比较每个模型的 API 服务商。 查看所有模型