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 (max) 和 Claude Opus 5 (xhigh) 是智能得分最高的模型，之后是 Claude Fable 5 (with fallback) 和 Claude Opus 5 (high)。
输出速度
Celeris-1 和 Mercury 2 是速度最快的模型，之后是 LFM2.5-VL-1.6B 和 Gemini 2.5 Flash-Lite。
延迟
Gemini 2.5 Flash-Lite 和 Command A+ 是延迟最低的模型，之后是 Granite 4.2 3B 和 Gemini 2.5 Flash。
每任务成本
Granite 4.2 3B 和 GPT-5.6 Luna (low) 的每任务成本最低，之后是 MiMo-V2.5 和 Llama 4 Scout。
上下文窗口
Llama 4 Scout 和 Grok 4.20 0309 支持最大的上下文窗口，之后是 Gemini 1.5 Pro (May) 和 Grok 4.1 Fast。
进一步分析 | ||||||||
|---|---|---|---|---|---|---|---|---|
Claude Opus 5 (max) | 1M | Anthropic | 63 | $2.34 | 55 | 42.34 | 51.35 | |
Claude Opus 5 (xhigh) | 1M | Anthropic | 63 | $1.80 | 54 | 32.67 | 41.97 | |
Claude Fable 5 (with fallback) | 1M | Anthropic | 62 | $3.14 | 70 | 96.50 | 103.65 | |
Claude Opus 5 (high) | 1M | Anthropic | 61 | $1.23 | 55 | 16.37 | 25.47 | |
GPT-5.6 Sol (max) | 1M | OpenAI | 61 | $1.01 | 72 | 122.11 | 129.07 | |
Grok 4.6 (high) | 500k | SpaceXAI | 61 | $0.94 | 57 | 43.13 | 51.84 | |
Grok 4.6 (xhigh) | 500k | SpaceXAI | 60 | $1.23 | 61 | 44.37 | 52.58 | |
Kimi K3 (max) | 1.05M |  Kimi | 60 | $0.84 | 39 | 7.02 | 71.11 | |
GLM-5.3 (max) | 1M |  Z AI | 60 | $0.68 | 85 | 1.57 | 30.87 | |
GPT-5.6 Sol (xhigh) | 1M | OpenAI | 59 | $0.67 | 72 | 50.12 | 57.02 | |
Grok 4.6 (medium) | 500k | SpaceXAI | 59 | $0.78 | 58 | 34.53 | 43.12 | |
Claude Opus 5 (medium) | 1M | Anthropic | 59 | $0.72 | 54 | 5.13 | 14.41 | |
Qwen3.8 Max | 1M |  Alibaba | 58 | $0.91 | 21 | 2.49 | 123.31 | |
Qwen3.8 2.4T A95B | 984k | Alibaba | 58 | $0.81 | 24 | 2.58 | 107.96 | |
GLM-5.3-Flash | 1M | Z AI | 57 | $0.09 | 50 | 1.47 | 51.24 | |
GPT-5.6 Sol (high) | 1M | OpenAI | 57 | $0.46 | 75 | 11.06 | 17.75 | |
Muse Spark 1.2 (xhigh) | 1.05M | Meta | 57 | $0.40 | -- | -- | -- | |
GPT-5.6 Terra (max) | 1M | OpenAI | 57 | $0.53 | 109 | 150.94 | 155.52 | |
Gemini 3.7 Flash (high) | 1M | Google | 56 | $0.40 | 345 | 11.97 | 13.42 | |
Grok 4.5 (high) | 500k | SpaceXAI | 56 | $0.43 | 56 | 14.21 | 23.12 | |
GPT-5.6 Sol (medium) | 1M | OpenAI | 56 | $0.31 | 73 | 4.69 | 11.55 | |
Claude Sonnet 5 (max) | 1M | Anthropic | 55 | $1.72 | 90 | 195.84 | 201.42 | |
Gemini 3.7 Flash (medium) | 1M | Google | 53 | $0.26 | 342 | 5.88 | 7.34 | |
DeepSeek V4 Pro 0813 (max) | 1M |  DeepSeek | 53 | $0.27 | 68 | 1.51 | 38.29 | |
GPT-5.6 Terra (xhigh) | 1M | OpenAI | 53 | $0.32 | 106 | 29.78 | 34.49 | |
GLM-5.2 (max) | 1M | Z AI | 53 | $0.44 | 69 | 1.57 | 37.63 | |
Claude Opus 5 (low) | 1M | Anthropic | 52 | $0.43 | 55 | 3.13 | 12.16 | |
GPT-5.6 Luna (max) | 1M | OpenAI | 52 | $0.05 | 131 | 110.84 | 114.66 | |
Qwen3.8 27B (xhigh) | 256k | Alibaba | 52 | $0.37 | 50 | 3.38 | 52.97 | |
DeepSeek V4 Flash 0731 (max) | 1M | DeepSeek | 52 | $0.11 | 119 | 1.12 | 22.07 | |
Grok 4.6 (low) | 500k | SpaceXAI | 52 | $0.25 | 56 | 7.24 | 16.09 | |
Gemini 3.6 Flash | 1M | Google | 52 | $0.34 | 199 | 16.68 | 19.20 | |
DeepSeek V4 Flash Vision (max) | 1M | DeepSeek | 51 | $0.12 | 119 | 1.14 | 22.09 | |
Gemini 3.7 Flash (low) | 1M | Google | 51 | $0.16 | 320 | 1.00 | 2.56 | |
GPT-5.6 Sol (low) | 1M | OpenAI | 51 | $0.19 | 75 | 2.99 | 9.69 | |
GPT-5.6 Terra (high) | 1M | OpenAI | 50 | $0.23 | 102 | 3.35 | 8.26 | |
GPT-5.6 Luna (xhigh) | 1M | OpenAI | 50 | $0.03 | 119 | 54.47 | 58.68 | |
Agnes 2.5 Pro Beta | 1M |  Sapiens AI | 49 | -- | 157 | 2.43 | 18.38 | |
Kimi K3 (low) | 1.05M | Kimi | 48 | $0.24 | 36 | 4.92 | 73.49 | |
Gemini 3.1 Pro Preview | 1M | Google | 48 | $0.33 | 115 | 33.04 | 37.38 | |
Motif 3 | 262k |  Motif Technologies | 47 | -- | -- | -- | -- | |
GPT-5.6 Luna (high) | 1M | OpenAI | 47 | $0.02 | 122 | 11.69 | 15.79 | |
GPT-5.6 Terra (medium) | 1M | OpenAI | 47 | $0.12 | 96 | 1.77 | 6.98 | |
Gemini 3.5 Flash (medium) | 1M | Google | 47* | -- | 196 | 15.75 | 18.30 | |
GPT-5.3 Codex (xhigh) | 400k | OpenAI | 46* | -- | 130 | 68.12 | 71.97 | |
MiniMax-M3 | 1M |  MiniMax | 45 | $0.14 | 118 | 1.34 | 22.55 | |
Motif 3 (Beta) | 262k | Motif Technologies | 45* | -- | -- | -- | -- | |
DeepSeek V4 Pro (max) | 1M | DeepSeek | 45 | $0.05 | 67 | 1.74 | 74.50 | |
Qwen3.8 27B (medium) | 256k | Alibaba | 44 | $0.28 | 53 | 3.49 | 50.86 | |
DeepSeek V4 Pro (high) | 1M | DeepSeek | 44 | $0.04 | 68 | 1.65 | 38.14 | |
Kimi K2.7 Code | 256k | Kimi | 43 | $0.22 | 45 | 2.96 | 64.27 | |
MiMo-V2.5-Pro | 1M | Xiaomi | 43 | $0.03 | 42 | 3.78 | 63.53 | |
Qwen3.8 27B (low) | 256k | Alibaba | 43 | $0.26 | 57 | 3.25 | 47.27 | |
Claude Sonnet 5 (Non-reasoning) | 1M | Anthropic | 43 | $0.42 | 64 | 1.70 | 9.54 | |
Inkling | 1M |  Thinking Machines | 42 | $0.34 | 50 | 3.18 | 53.01 | |
Hy3 | 256k |  Tencent | 42 | $0.04 | 68 | 2.84 | 39.49 | |
GPT-5.6 Sol (Non-reasoning) | 1M | OpenAI | 42 | $0.19 | 72 | 1.43 | 8.37 | |
Nex-N2-Pro | 262k |  Nex AGI | 42 | -- | 130 | 1.56 | 20.76 | |
Solar Pro 4 | 512k |  Upstage | 42 | $0.30 | 57 | 2.16 | 45.84 | |
GPT-5.6 Terra (low) | 1M | OpenAI | 41 | $0.10 | 102 | 1.52 | 6.43 | |
Inkling Small | 1M | Thinking Machines | 41 | $0.07 | 116 | 2.31 | 23.78 | |
JT-4.1 Flash 236B A21B | 256k |  China Mobile | 40* | -- | -- | -- | -- | |
Agnes 2.5 Pro Alpha | 1M | Sapiens AI | 40 | -- | 129 | 2.05 | 21.45 | |
Qwen3.7 Plus | 1M | Alibaba | 39 | $0.17 | 56 | 2.12 | 46.57 | |
GPT-5.6 Luna (medium) | 1M | OpenAI | 39 | $0.01 | 120 | 2.73 | 6.89 | |
Nemotron 3 Ultra | 262k |  NVIDIA | 38 | $0.38 | 176 | 2.09 | 17.84 | |
MiMo-V2.5 | 1M | Xiaomi | 38 | $0.01 | 69 | 3.07 | 39.19 | |
Ling 3.0 Flash | 262k |  InclusionAI | 38 | $0.04 | 388 | 2.50 | 8.94 | |
Qwen3.6 27B | 262k | Alibaba | 38 | $0.29 | 49 | 3.83 | 128.79 | |
Gemini 3.5 Flash-Lite | 1M | Google | 37 | $0.10 | 369 | 8.76 | 10.11 | |
Solar Open2 250B | 1.05M | Upstage | 37 | -- | -- | -- | -- | |
MiMo-V2-Omni-0327 | 256k | Xiaomi | 37* | -- | -- | -- | -- | |
Grok 4.3 (medium) | 1M | SpaceXAI | 37* | -- | 121 | 9.07 | 13.21 | |
Grok 4.3 (low) | 1M | SpaceXAI | 36* | -- | 120 | 6.44 | 10.61 | |
MiMo-V2-Omni | 256k | Xiaomi | 36* | -- | -- | -- | -- | |
Gemini 3.5 Flash (minimal) | 1M | Google | 36* | -- | 183 | 0.83 | 3.57 | |
Claude Sonnet 4.6 (Non-reasoning, Low Effort) | 1M | Anthropic | 35* | -- | 48 | 1.56 | 12.05 | |
Muse Glimmer (high) | 131k | Meta | 35 | $0.06 | 107 | 0.79 | 24.07 | |
A.X-K2 | 262k |  SK Telecom | 35 | -- | -- | -- | -- | |
GLM-5.2 | 1M | Z AI | 35 | -- | 102 | 2.02 | 6.93 | |
Qwen3.8 27B | 256k | Alibaba | 35 | $0.43 | 57 | 3.54 | 12.36 | |
GPT-5.6 Terra (Non-reasoning) | 1M | OpenAI | 35 | $0.10 | 103 | 0.89 | 5.72 | |
Qwen3.5 397B A17B | 262k | Alibaba | 34 | $0.36 | 94 | 2.29 | 41.36 | |
MiMo-V2-Flash (Feb 2026) | 256k | Xiaomi | 34* | -- | -- | -- | -- | |
G9v3-39A5B | 131k |  AI9Stars | 34 | $0.00 | -- | -- | -- | |
LongCat 2.0 | 1M |  LongCat | 34 | $0.12 | 41 | 2.80 | 64.42 | |
GPT-5.6 Luna (low) | 1M | OpenAI | 34 | $0.01 | 112 | 1.62 | 6.10 | |
KAT-Coder-Pro V2 | 256k |  KwaiKAT | 34 | -- | 101 | 1.41 | 6.36 | |
Qwen3.5 122B A10B | 262k | Alibaba | 33 | $0.25 | 134 | 2.33 | 21.04 | |
Qwen3.5 397B A17B | 262k | Alibaba | 33* | -- | 86 | 2.31 | 8.10 | |
Qwen3.6 35B A3B | 262k | Alibaba | 32 | $0.29 | 122 | 2.11 | 50.24 | |
DeepSeek V4 Pro | 1M | DeepSeek | 32* | -- | 70 | 1.63 | 8.80 | |
Ring-2.6-1T | 262k | InclusionAI | 32 | $0.36 | 129 | 3.91 | 23.30 | |
Qwen3.5 Omni Plus | 256k | Alibaba | 31* | -- | 131 | 2.28 | 6.09 | |
Qwen3.6 27B | 262k | Alibaba | 31 | $0.40 | 57 | 3.81 | 12.51 | |
o3 | 200k | OpenAI | 31* | -- | 119 | 8.85 | 13.06 | |
K-EXAONE 2.0 | 262k |  LG AI Research | 31 | -- | -- | -- | -- | |
Step 3.7 Flash | 262k |  StepFun | 31 | $0.09 | 92 | 2.84 | 30.11 | |
Mistral Medium 3.5 | 256k |  Mistral | 30 | $0.41 | 141 | 2.26 | 20.01 | |
Claude 4.5 Haiku | 200k | Anthropic | 30 | $0.22 | 128 | 17.25 | 21.16 | |
Gemma 4 31B | 256k | Google | 30 | $0.00 | 35 | 1.06 | 64.47 | |
DeepSeek V4 Flash | 1M | DeepSeek | 29* | -- | -- | -- | -- | |
GPT-5.5 Instant (June 2026) | 400k | OpenAI | 29 | $0.54 | 147 | 1.16 | 18.11 | |
JT-35B-Flash | 256k | China Mobile | 29* | -- | -- | -- | -- | |
MiMo-V2.5-Pro | 1M | Xiaomi | 28* | -- | 39 | 6.63 | 19.59 | |
Qwen3.5 122B A10B | 262k | Alibaba | 28 | $0.20 | 146 | 2.32 | 5.75 | |
GPT-5.6 Luna (Non-reasoning) | 1M | OpenAI | 27 | $0.01 | 110 | 0.86 | 5.41 | |
Doubao Seed Code | 256k |  ByteDance Seed | 26* | -- | -- | -- | -- | |
Gemma 4 26B A4B | 256k | Google | 26 | $0.04 | -- | -- | -- | |
Nemotron 3 Super | 1M | NVIDIA | 26 | $0.23 | 143 | 1.80 | 19.32 | |
MiMo-V2-Flash | 256k | Xiaomi | 25* | -- | -- | -- | -- | |
Grok 4.3 (Non-reasoning) | 1M | SpaceXAI | 25 | $0.29 | 124 | 0.75 | 4.77 | |
Qwen3.6 35B A3B | 262k | Alibaba | 25 | $0.63 | 139 | 2.03 | 5.63 | |
Ling 3.0 Tiny | 262k | InclusionAI | 25 | $0.00 | 154 | 2.70 | 18.96 | |
Qwen3.5 35B A3B | 262k | Alibaba | 24 | $0.24 | 169 | 2.04 | 5.00 | |
Claude 4.5 Haiku | 200k | Anthropic | 24* | -- | 98 | 0.77 | 5.86 | |
gpt-oss-120b (high) | 131k | OpenAI | 24 | $0.07 | 160 | 0.88 | 16.48 | |
Nemotron 3.5 Lightning | 1M | NVIDIA | 24 | $0.08 | 301 | 1.03 | 9.32 | |
Command A+ | 192k | Cohere | 23 | $0.00 | 254 | 0.38 | 10.23 | |
K-EXAONE | 256k | LG AI Research | 22* | -- | -- | -- | -- | |
ERNIE 5.0 Thinking Preview | 128k |  Baidu | 22* | -- | -- | -- | -- | |
Gemma 4 31B | 256k | Google | 22 | $0.04 | 84 | 1.35 | 7.31 | |
Gemma 4 12B | 256k | Google | 22* | -- | 110 | 2.27 | 25.07 | |
Nova 2.0 Pro Preview (medium) | 256k |  Amazon | 22* | -- | 119 | 16.04 | 37.11 | |
Mercury 2 | 128k | Inception | 22 | $0.08 | 1,005 | 3.63 | 4.13 | |
Qwen3.5 9B | 262k | Alibaba | 22 | $0.24 | 77 | 2.15 | 34.76 | |
Qwen3 Coder Next | 256k | Alibaba | 21 | $0.34 | 113 | 1.23 | 5.64 | |
Nova 2.0 Omni (medium) | 1M | Amazon | 21* | -- | -- | -- | -- | |
Apriel-v1.6-15B-Thinker | 128k |  ServiceNow | 21* | -- | -- | -- | -- | |
Nova 2.0 Lite (high) | 1M | Amazon | 21* | -- | 161 | 16.59 | 32.09 | |
Qwen3.5 9B | 262k | Alibaba | 21* | -- | 95 | 0.83 | 6.07 | |
EXAONE 4.5 33B | 262k | LG AI Research | 21* | -- | -- | -- | -- | |
Gemma 4 26B A4B | 256k | Google | 20* | -- | 72 | 1.25 | 8.19 | |
Qwen3.5 4B | 262k | Alibaba | 20* | -- | 43 | 0.66 | 58.64 | |
North Mini Code | 256k | Cohere | 20 | $0.00 | 35 | 0.49 | 71.19 | |
Nova 2.0 Pro Preview (low) | 256k | Amazon | 20* | -- | 120 | 11.20 | 32.04 | |
Mistral Small 4 | 256k | Mistral | 20 | $0.10 | 168 | 0.84 | 15.73 | |
Granite 4.2 8B | 131k | IBM | 20 | $0.02 | 139 | 0.63 | 18.65 | |
Devstral 2 | 256k | Mistral | 19 | $0.00 | 52 | 1.69 | 11.34 | |
Nova 2.0 Lite (medium) | 1M | Amazon | 19* | -- | 186 | 16.46 | 29.89 | |
Qwen3.5 Omni Flash | 256k | Alibaba | 19* | -- | 253 | 1.82 | 3.80 | |
JT-MINI | 128k | China Mobile | 19* | -- | -- | -- | -- | |
Trinity Large Thinking | 512k |  Arcee AI | 19 | $0.16 | 171 | 1.16 | 15.73 | |
HyperNova 60B 2605 | 131k |  Multiverse Computing | 18 | $0.02 | 366 | 0.81 | 7.64 | |
Magistral Medium 1.2 | 128k | Mistral | 18 | -- | -- | -- | -- | |
Nova 2.0 Lite (low) | 1M | Amazon | 18* | -- | 160 | 8.89 | 24.56 | |
Nemotron Cascade 2 30B A3B | 1M | NVIDIA | 18* | -- | -- | -- | -- | |
Devstral Small 2 | 256k | Mistral | 18 | $0.00 | 134 | 2.20 | 5.93 | |
K2 Think V2 | 262k |  MBZUAI Institute of Foundation Models | 17* | -- | -- | -- | -- | |
LongCat Flash Lite | 256k | LongCat | 17* | -- | -- | -- | -- | |
HyperCLOVA X SEED Think (32B) | 128k |  Naver | 17* | -- | -- | -- | -- | |
K-EXAONE | 256k | LG AI Research | 17* | -- | -- | -- | -- | |
Qwen3 Next 80B A3B | 262k | Alibaba | 17 | $0.04 | 190 | 2.26 | 15.41 | |
Nova 2.0 Omni (low) | 1M | Amazon | 17* | -- | -- | -- | -- | |
Mi:dm K 2.5 Pro | 128k |  Korea Telecom | 17* | -- | -- | -- | -- | |
G9v3-3B | 131k | AI9Stars | 16* | -- | -- | -- | -- | |
Qwen3.5 4B | 262k | Alibaba | 16* | -- | 31 | 0.73 | 16.67 | |
Mistral Large 3 | 256k | Mistral | 16 | $0.08 | 37 | 1.25 | 14.88 | |
INTELLECT-3 | 131k |  Prime Intellect | 16* | -- | -- | -- | -- | |
Solar Open 100B | 128k | Upstage | 15* | -- | -- | -- | -- | |
gpt-oss-20b (high) | 131k | OpenAI | 15 | $0.02 | 121 | 1.06 | 21.80 | |
Nemotron 3 Nano Omni 30B A3B | 256k | NVIDIA | 15* | -- | 327 | 0.99 | 8.63 | |
gpt-oss-120b (low) | 131k | OpenAI | 15 | $0.02 | 141 | 0.85 | 18.62 | |
Nemotron 3 Nano | 1M | NVIDIA | 15 | $0.02 | 232 | 1.16 | 11.95 | |
Solar Pro 3 | 128k | Upstage | 14 | $0.13 | 154 | 2.39 | 18.67 | |
Llama 4 Maverick | 1M | Meta | 14 | $0.03 | 100 | 0.90 | 5.90 | |
Nova 2.0 Pro Preview | 256k | Amazon | 14* | -- | 106 | 1.03 | 5.73 | |
gpt-oss-20b (low) | 131k | OpenAI | 14* | -- | 151 | 1.18 | 17.73 | |
Granite 4.2 3B | 131k | IBM | 14 | $0.0047 | 207 | 0.46 | 12.53 | |
K2-V2 (high) | 512k | MBZUAI Institute of Foundation Models | 14* | -- | -- | -- | -- | |
Qwen3 Next 80B A3B | 262k | Alibaba | 14* | -- | 180 | 2.12 | 4.91 | |
DiffusionGemma 26B A4B | 256k | Google | 13* | -- | -- | -- | -- | |
Gemma 4 12B (Non-reasoning) | 262k | Google | 13* | -- | 116 | 2.29 | 6.60 | |
Motif-2-12.7B | 128k | Motif Technologies | 13* | -- | -- | -- | -- | |
Nova Premier | 1M | Amazon | 13* | -- | 32 | 2.81 | 18.62 | |
K2-V2 (medium) | 512k | MBZUAI Institute of Foundation Models | 12* | -- | -- | -- | -- | |
Llama Nemotron Super 49B v1.5 | 128k | NVIDIA | 12* | -- | 118 | 6.51 | 27.65 | |
Celeris-1 | 131k | Celeris | 12 | $0.26 | 1,595 | 0.57 | 0.88 | |
Mistral Small 4 | 256k | Mistral | 12* | -- | 147 | 0.85 | 4.26 | |
Tri-21B-Think | 32k |  Trillion Labs | 12* | -- | -- | -- | -- | |
Gemma 4 E4B | 128k | Google | 12* | -- | 48 | 0.79 | 52.36 | |
MiniCPM5-1B | 128k |  OpenBMB | 12* | -- | -- | -- | -- | |
Sarvam 105B (high) | 128k |  Sarvam | 12* | -- | -- | -- | -- | |
Nova 2.0 Lite | 1M | Amazon | 12* | -- | 168 | 1.09 | 4.06 | |
MiniCPM5-1B | 128k | OpenBMB | 12* | -- | -- | -- | -- | |
Magistral Small 1.2 | 128k | Mistral | 11 | $0.30 | -- | -- | -- | |
Ministral 3 14B | 256k | Mistral | 11 | $0.16 | 95 | 0.93 | 6.21 | |
Nanbeige4.1-3B | 256k |  Nanbeige | 11* | -- | -- | -- | -- | |
LFM2.5-2.6B | 128k | Liquid AI | 11 | $0.00 | -- | -- | -- | |
EXAONE 4.0 32B | 131k | LG AI Research | 11* | -- | -- | -- | -- | |
Nova 2.0 Omni | 1M | Amazon | 10* | -- | -- | -- | -- | |
Llama 4 Scout | 10M | Meta | 10 | $0.01 | 132 | 0.78 | 4.56 | |
Hermes 4 70B | 128k |  Nous Research | 10* | -- | 81 | 1.48 | 32.51 | |
Falcon-H1R-7B | 256k |  TII UAE | 10* | -- | -- | -- | -- | |
Gemma 4 E2B | 128k | Google | 10* | -- | -- | -- | -- | |
Qwen3 Omni 30B A3B | 65.5k | Alibaba | 10* | -- | 101 | 1.89 | 26.58 | |
Step3 VL 10B | 65.5k | StepFun | 9* | -- | -- | -- | -- | |
Llama 3.3 70B | 128k | Meta | 9* | -- | 82 | 1.70 | 7.79 | |
Ministral 3 8B | 256k | Mistral | 9 | $0.18 | 125 | 0.76 | 4.76 | |
Llama Nemotron Ultra | 128k | NVIDIA | 9* | -- | 52 | 2.39 | 50.04 | |
ERNIE 4.5 300B A47B | 131k | Baidu | 9* | -- | -- | -- | -- | |
Hermes 4 405B | 128k | Nous Research | 9* | -- | 34 | 2.51 | 76.23 | |
NVIDIA Nemotron Nano 12B v2 VL | 128k | NVIDIA | 9* | -- | 104 | 8.24 | 32.19 | |
Gemma 4 E4B | 128k | Google | 9* | -- | 53 | 0.73 | 10.16 | |
Granite 4.1 30B | 131k | IBM | 9* | -- | -- | -- | -- | |
NVIDIA Nemotron Nano 9B V2 | 131k | NVIDIA | 9* | -- | 214 | 3.23 | 14.93 | |
Hermes 4 405B | 128k | Nous Research | 9* | -- | 32 | 2.39 | 17.94 | |
Nemotron 3 Nano 4B | 262k | NVIDIA | 9* | -- | -- | -- | -- | |
Llama Nemotron Super 49B v1.5 | 128k | NVIDIA | 9* | -- | 110 | 3.76 | 8.32 | |
K2-V2 (low) | 512k | MBZUAI Institute of Foundation Models | 8* | -- | -- | -- | -- | |
Kimi Linear 48B A3B Instruct | 1M | Kimi | 8* | -- | -- | -- | -- | |
Llama 3.1 405B | 128k | Meta | 8* | -- | -- | -- | -- | |
LFM2.5-8B-A1B | 32.8k | Liquid AI | 8* | -- | 338 | 1.43 | 8.83 | |
Ring-flash-2.0 | 128k | InclusionAI | 8* | -- | -- | -- | -- | |
Olmo 3.1 32B Think | 65.5k |  Allen Institute for AI | 8* | -- | -- | -- | -- | |
Command A | 256k | Cohere | 7* | -- | 63 | 1.56 | 9.48 | |
Qwen3.5 2B | 262k | Alibaba | 7* | -- | -- | -- | -- | |
Llama 3.1 Nemotron 70B | 128k | NVIDIA | 7* | -- | 98 | 8.57 | 13.66 | |
Nemotron 3 Nano | 1M | NVIDIA | 7* | -- | 223 | 0.95 | 3.19 | |
NVIDIA Nemotron Nano 9B V2 | 131k | NVIDIA | 7* | -- | 169 | 1.27 | 4.23 | |
Ministral 3 3B | 256k | Mistral | 7 | $0.13 | 191 | 0.62 | 3.24 | |
Hermes 4 70B | 128k | Nous Research | 7* | -- | 82 | 1.39 | 7.51 | |
Granite 4.1 8B | 131k | IBM | 6* | -- | 120 | 0.78 | 4.94 | |
Sarvam 30B (high) | 65.5k | Sarvam | 6* | -- | -- | -- | -- | |
Olmo 3.1 32B Instruct | 65.5k | Allen Institute for AI | 6* | -- | -- | -- | -- | |
Gemma 4 E2B | 128k | Google | 6* | -- | -- | -- | -- | |
R1 1776 | 128k |  Perplexity | 6* | -- | -- | -- | -- | |
Llama 3.2 90B (Vision) | 128k | Meta | 6* | -- | -- | -- | -- | |
Phi-4 Mini | 128k |  Microsoft | 6* | -- | 43 | 0.88 | 12.37 | |
EXAONE 4.0 32B | 131k | LG AI Research | 6* | -- | -- | -- | -- | |
Qwen3.5 2B | 262k | Alibaba | 5* | -- | -- | -- | -- | |
Qwen3.5 0.8B | 262k | Alibaba | 5* | -- | -- | -- | -- | |
DeepHermes 3 - Mistral 24B | 32k | Nous Research | 5* | -- | -- | -- | -- | |
Jamba 1.7 Large | 256k |  AI21 Labs | 5* | -- | -- | -- | -- | |
Granite 4.0 H Small | 128k | IBM | 5* | -- | 22 | 23.18 | 45.94 | |
Qwen3 Omni 30B A3B | 65.5k | Alibaba | 5* | -- | 94 | 1.93 | 7.27 | |
LFM2 24B A2B | 32.8k | Liquid AI | 5* | -- | -- | -- | -- | |
Phi-4 | 16k | Microsoft | 5* | -- | 41 | 2.47 | 14.72 | |
Nova Micro | 130k | Amazon | 4* | -- | 300 | 0.85 | 2.52 | |
Granite 4.1 3B | 131k | IBM | 4* | -- | -- | -- | -- | |
NVIDIA Nemotron Nano 12B v2 VL | 128k | NVIDIA | 4* | -- | 162 | 1.70 | 4.78 | |
Phi-4 Multimodal | 128k | Microsoft | 4* | -- | 18 | 0.82 | 29.36 | |
MiniCPM-V 4.6 1.3B | 262k | OpenBMB | 4* | -- | -- | -- | -- | |
Jamba Reasoning 3B | 262k | AI21 Labs | 4* | -- | -- | -- | -- | |
Reka Flash 3 | 128k |  Reka AI | 4* | -- | -- | -- | -- | |
Olmo 3 7B Think | 65.5k | Allen Institute for AI | 4* | -- | -- | -- | -- | |
Molmo 7B-D | 4.1k | Allen Institute for AI | 3* | -- | -- | -- | -- | |
Llama 3.2 11B (Vision) | 128k | Meta | 3* | -- | 50 | 0.98 | 10.89 | |
Qwen3.5 0.8B | 262k | Alibaba | 3* | -- | -- | -- | -- | |
Exaone 4.0 1.2B | 64k | LG AI Research | 3* | -- | -- | -- | -- | |
Olmo 3 7B | 65.5k | Allen Institute for AI | 2* | -- | -- | -- | -- | |
Exaone 4.0 1.2B | 64k | LG AI Research | 2* | -- | -- | -- | -- | |
LFM2.5-1.2B-Thinking | 32k | Liquid AI | 2* | -- | -- | -- | -- | |
Jamba 1.7 Mini | 258k | AI21 Labs | 2* | -- | -- | -- | -- | |
LFM2 2.6B | 32.8k | Liquid AI | 2* | -- | -- | -- | -- | |
LFM2.5-1.2B-Instruct | 32k | Liquid AI | 2* | -- | -- | -- | -- | |
Granite 4.0 H 1B | 128k | IBM | 2* | -- | -- | -- | -- | |
Gemma 3 270M | 32k | Google | 2* | -- | -- | -- | -- | |
Apertus 70B Instruct | 65.5k |  Swiss AI Initiative | 2* | -- | -- | -- | -- | |
Granite 4.0 Micro | 128k | IBM | 2* | -- | -- | -- | -- | |
DeepHermes 3 - Llama-3.1 8B | 128k | Nous Research | 2* | -- | -- | -- | -- | |
Granite 4.0 1B | 128k | IBM | 2* | -- | -- | -- | -- | |
Molmo2-8B | 36.9k | Allen Institute for AI | 2* | -- | -- | -- | -- | |
LFM2 8B A1B | 32.8k | Liquid AI | 1* | -- | -- | -- | -- | |
LFM2.5-VL-1.6B | 32k | Liquid AI | 1* | -- | 400 | 1.39 | 2.65 | |
Granite 4.0 350M | 32.8k | IBM | 1* | -- | -- | -- | -- | |
Tiny Aya Global | 8.19k | Cohere | 1* | -- | -- | -- | -- | |
Apertus 8B Instruct | 65.5k | Swiss AI Initiative | 1* | -- | -- | -- | -- | |
Granite 4.0 H 350M | 32.8k | IBM | 1* | -- | -- | -- | -- | |
Claude Sonnet 5 (low) | 1M | Anthropic | -- | -- | 64 | 2.09 | 9.92 | |
EXAONE 4.5 33B | 262k | LG AI Research | -- | -- | -- | -- | -- | |
Claude Sonnet 5 (xhigh) | 1M | Anthropic | -- | -- | 69 | 24.02 | 31.26 | |
Gemini 3 Deep Think | 128k | Google | -- | -- | -- | -- | -- | |
Claude Sonnet 5 (high) | 1M | Anthropic | -- | -- | 68 | 11.83 | 19.20 | |
GPT-5.5 Pro (xhigh) | 922k | OpenAI | -- | -- | -- | -- | -- | |
Cogito v2.1 | 128k |  Deep Cogito | -- | -- | -- | -- | -- | |
Claude Sonnet 5 (medium) | 1M | Anthropic | -- | -- | 65 | 3.09 | 10.76 | |
关键定义
常见问题
在参与排名的 175 个模型中，Claude Opus 5 (Adaptive Reasoning, Max Effort) 目前以 63 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 (Adaptive Reasoning, Max Effort)（63）, 2. Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)（63）, 3. Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)（62）, 4. Claude Opus 5 (Adaptive Reasoning, High Effort)（61）, 5. GPT-5.6 Sol (max)（61）。
Celeris-1 最快，输出速度为每秒 1,594.8 个 token，其次是 Mercury 2（1,004.5 t/s）和 LFM2.5-VL-1.6B（399.8 t/s）。
Granite 4.2 3B 的每项 Intelligence Index 任务成本最低，为 $0.0047，其次是 GPT-5.6 Luna (low)（$0.01）和 MiMo-V2.5（$0.01）。
Kimi K3 (max) 是排名最高的开放权重模型，Intelligence Index 得分为 60。排行榜上的 175 个模型中，有 95 个开放权重模型。
按 Intelligence Index 排名，顶尖开放权重模型是：1. Kimi K3 (max)（60）, 2. Qwen3.8 2.4T A95B（58）, 3. GLM-5.3-Flash（57）。
在 141 个推理模型中，Claude Opus 5 (Adaptive Reasoning, Max Effort) 以 63 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