The AI tools index that doesn't waste your time.不浪费你时间的 AI 工具索引。
LynxModel

AI Model Intelligence RankingsAI 模型智能排行

60 frontier models, one transparent 100-point Intelligence Score across 11 public benchmarks (reasoning, coding, knowledge, math, agent, human preference). Every score links to its source. Data refreshed 2026-10-20. Click any column to sort.

60 个前沿模型,一套透明的百分制智能评分,覆盖 11 个公开 benchmark(推理/代码/知识/数学/智能体/人类偏好)。每个分数都带来源链接。数据更新于 2026-10-20。点表头可排序。

Find my model帮我选型 Pricing & value价格与性价比 Head-to-head模型对战 Image arena图片生成榜 Video arena视频生成榜 How the score works评分方法 Benchmark explorerBenchmark 百科

Rea = Reasoning推理 Cod = Coding代码 Kno = Knowledge知识 Mat = Math数学 Agt = Agent智能体 Pre = Preference偏好

  1. Claude Fable 5 — 95.4
  2. GPT-5.6 Sol — 94.3
  3. DeepSeek V4 Flash 0731 — 91.1
  4. GPT-5.6 Terra — 89.7
  5. Gemini 3.1 Pro Preview — 89.6
  6. GPT-5.6 Sol Pro — 88.8
  7. GLM 5.2 — 88.7
  8. Kimi K3 — 87.9
  9. DeepSeek V4 Pro 0423 — 87.4
  10. Claude Opus 4.8 — 87
⚖ Adjust the weights — the ranking recomputes live⚖ 调权重——排名实时重算
30
25
15
15
10
5
Same formula as the published score: weighted average over the dimensions a model actually has, renormalized by the weights present. Below 30% coverage a model shows no score; below 60% it carries a low-data badge.与公开评分同一公式:按模型实际拥有的维度加权平均、按出席权重重归一;覆盖<30% 不出分,<60% 打"数据少"徽标。

Intelligence vs price — the value map智能分 vs 价格——性价比地图

Each dot is a scored model: higher = smarter, further left = cheaper. Top-left is the sweet spot. Color = provider, dashed lines = medians.每个点是一个有评分的模型:越往上越聪明,越往左越便宜,左上角是甜区。颜色=厂商,虚线=中位数。 Click a dot for the model page. 点圆点进模型页。

9880634527$0.03$0.10$0.32$1.0$3.2$10$32blended $/1M tokens (log) →Intelligence Score →Claude Fable 5 (Anthropic) · Score 95.4 · $10/$50 per 1MGPT-5.6 Sol (OpenAI) · Score 94.3 · $4/$20 per 1MDeepSeek V4 Flash 0731 (DeepSeek) · Score 91.1 · $0.01/$0.64 per 1MGPT-5.6 Terra (OpenAI) · Score 89.7 · $2/$12 per 1MGemini 3.1 Pro Preview (Google) · Score 89.6 · $2/$12 per 1MGPT-5.6 Sol Pro (OpenAI) · Score 88.8 · $4/$20 per 1MGLM 5.2 (Z.ai) · Score 88.7 · $0.1/$7 per 1MKimi K3 (Moonshot AI) · Score 87.9 · $0.3/$14 per 1MDeepSeek V4 Pro 0423 (DeepSeek) · Score 87.4 · $0.3717/$0.7433 per 1MClaude Opus 4.8 (Anthropic) · Score 87 · $5/$25 per 1MQwen3.8 Max Prime (Qwen) · Score 86.4 · $4/$12 per 1MQwen3.7 Max (Qwen) · Score 85.2 · $1.475/$4.425 per 1MGrok 4.20 (xAI) · Score 85.1 · $1.25/$2.5 per 1MClaude Opus 5 (Anthropic) · Score 85 · $5/$25 per 1MGPT-5.6 Luna (OpenAI) · Score 84.7 · $0.2/$1.2 per 1MHy3 (Tencent) · Score 84 · $0.0825/$0.33 per 1MGemini 3.5 Flash (Google) · Score 83.6 · $1.5/$9 per 1MDeepSeek V4 Flash 0423 (DeepSeek) · Score 83.4 · $0.091/$0.182 per 1MQwen3.7 Plus (Qwen) · Score 83.4 · $0.32/$1.28 per 1MMuse Spark 1.2 (meta) · Score 83 · $1.25/$4.25 per 1MGPT-5.5 (OpenAI) · Score 82 · $5/$30 per 1MQwen3.7 Flash (Qwen) · Score 82 · $0.03/$0.13 per 1MMiniMax M3 (MiniMax) · Score 81.7 · $0.3/$1.2 per 1MStep 3.7 Flash (StepFun) · Score 81.4 · $0.2/$1.15 per 1MClaude Sonnet 5 (Anthropic) · Score 80.9 · $2/$10 per 1MGemini 3.6 Flash (Google) · Score 80.1 · $0.75/$3.75 per 1MKimi K2.7 Code (Moonshot AI) · Score 80.1 · $0.6712/$3.35 per 1MGrok 4.5 (xAI) · Score 79.5 · $2/$6 per 1MSeed-2.0-Mini (ByteDance Seed) · Score 77.7 · $0.1/$0.4 per 1MNemotron 3 Ultra (NVIDIA) · Score 76.4 · $0.6/$2.4 per 1MMiniMax M2.7 (MiniMax) · Score 75.7 · $0.3/$1.2 per 1MMiMo-V2.5-Pro (Xiaomi) · Score 75 · $0.435/$0.87 per 1MGrok 4.3 (xAI) · Score 74.3 · $1.25/$2.5 per 1MMuse Spark 1.1 (meta) · Score 71.8 · $1.25/$4.25 per 1MGPT-5.6 Luna Pro (OpenAI) · Score 71.5 · $0.2/$1.2 per 1MKimi K2 Thinking (Moonshot AI) · Score 71 · $0.6/$2.5 per 1MDeepSeek V3.2 (DeepSeek) · Score 70.9 · $0.259/$0.8 per 1MGLM 5.1 (Z.ai) · Score 70.1 · $0.966/$3.036 per 1MMistral Medium 3.5 (Mistral AI) · Score 65.4 · $1.5/$7.5 per 1MSeed-2.0-Lite (ByteDance Seed) · Score 65.3 · $0.25/$2 per 1Mgpt-oss-120b (OpenAI) · Score 64.9 · $0.15/$0.6 per 1MGemini 3.5 Flash Lite (Google) · Score 63.4 · $0.3/$2.5 per 1MGemini 3.1 Flash Lite (Google) · Score 58.6 · $0.25/$1.5 per 1MMercury 2 (Inception) · Score 57.2 · $0.25/$0.75 per 1MClaude Haiku 4.5 (Anthropic) · Score 54.2 · $1/$5 per 1MSonar Pro (Perplexity) · Score 52.5 · $3/$15 per 1MCommand A (Cohere) · Score 50.7 · $2.5/$10 per 1MMistral Small 4 (Mistral AI) · Score 50.2 · $0.15/$0.6 per 1MMistral Large (Mistral AI) · Score 49.9 · $2/$6 per 1MCommand R+ (08-2024) (Cohere) · Score 42.8 · $2.5/$10 per 1MLlama 4 Maverick (Meta) · Score 42.5 · $0.1875/$0.6525 per 1MNova Pro 1.0 (Amazon) · Score 39.1 · $0.8/$3.2 per 1MPhi 4 (Microsoft) · Score 38.5 · $0.07/$0.14 per 1MLlama 4 Scout (Meta) · Score 34.5 · $0.18/$0.59 per 1MNova Premier 1.0 (Amazon) · Score 30.6 · $2.5/$12.5 per 1MClaude Fable 5GPT-5.6 SolDeepSeek V4 Flash 0731GPT-5.6 TerraGemini 3.1 Pro Preview
#Model模型Score总分 Rea推理 Cod代码 Kno知识 Mat数学 Agt智能体 Pre偏好 Ctx上下文 $ in/1M输入$/1M $ out/1M输出$/1M
1Claude Fable 5(max)Anthropic95.495.796.298.896.999.470.81M$10$50
2GPT-5.6 Sol(max)OpenAI94.390.496.396.29692.91001.1M$4.00$20
3DeepSeek V4 Flash 0731DeepSeek91.187.294.395.193.1——1M$0.010$0.640
4GPT-5.6 Terra(max)OpenAI89.788.888.893.693.987.174.81.1M$2.00$12
5Gemini 3.1 Pro Preview(high)Google89.690.584.610096.385.765.81M$2.00$12
6GPT-5.6 Sol Pro(max)OpenAI88.8 low data数据少91.9—90.7——64.51.1M$4.00$20
7GLM 5.2(max)Z.ai88.795.4739699.3100621M$0.100$7.00
8Kimi K3(always-on常开)Moonshot AI87.987.597.79570.793.866.51M$0.300$14
9DeepSeek V4 Pro 0423(max)DeepSeek87.489.285.994.585.597.158.81M$0.372$0.743
10Claude Opus 4.8(max)Anthropic8789.49096.875.986.262.81M$5.00$25
11Qwen3.8 Max PrimeQwen86.497.493.295.754.972.5671M$4.00$12
12Qwen3.7 Max(on/off开/关)Qwen85.29585.796.857.295.662.81M$1.48$4.42
13Grok 4.20(high)xAI85.191.875.893.282.297.4632M$1.25$2.50
14Claude Opus 5(max)Anthropic8596.485.988.567.27966.51M$5.00$25
15GPT-5.6 Luna(max)OpenAI84.780.186.392.992.4—57.21.1M$0.200$1.20
16Hy3Tencent84 low data数据少88.3————58.5262K$0.083$0.330
17Gemini 3.5 Flash(high)Google83.683.385.196.770.996.263.51M$1.50$9.00
18DeepSeek V4 Flash 0423(max)DeepSeek83.473.788.493.19796.5531M$0.091$0.182
19Qwen3.7 Plus(on/off开/关)Qwen83.477.586.995.685.793.858.81M$0.320$1.28
20Muse Spark 1.2meta83 low data数据少82.787.5———68.81M$1.25$4.25
21GPT-5.5(max)OpenAI8280.589.19385.555.464.81.1M$5.00$30
22Qwen3.7 Flash(on/off开/关)Qwen8278.587.396.467.3—631M$0.030$0.130
23MiniMax M3MiniMax81.780.18290.983.982.255.21M$0.300$1.20
24Step 3.7 FlashStepFun81.486.474.3—97.399.443262K$0.200$1.15
25Claude Sonnet 5(high)Anthropic80.989.188.294.549.175.559.71M$2.00$10
26Gemini 3.6 Flash(high)Google80.184.591.759.877.7—65.31M$0.750$3.75
27Kimi K2.7 CodeMoonshot AI80.169.38291.410090.959.5262K$0.671$3.35
28Grok 4.5(high)xAI79.583.788.696.358.73861.5500K$2.00$6.00
29Seed-2.0-MiniByteDance Seed77.7 low data数据少82.769.8—87——262K$0.100$0.400
30Nemotron 3 UltraNVIDIA76.468.781.593.7——51262K$0.600$2.40
31MiniMax M2.7MiniMax75.771.177.788.678.37748.3205K$0.300$1.20
32MiMo-V2.5-ProXiaomi75 low data数据少87.4——37.3—61.31.1M$0.435$0.870
33Grok 4.3(high)xAI74.377.770.892.753.498.654.71M$1.25$2.50
34Muse Spark 1.1meta71.886.182.9—39.674.866.51M$1.25$4.25
35GPT-5.6 Luna Pro(max)OpenAI71.5 low data数据少85.422———56.71.1M$0.200$1.20
36Kimi K2 Thinking(always-on常开)Moonshot AI7165.569.894.661.978.951.8262K$0.600$2.50
37DeepSeek V3.2(on/off开/关)DeepSeek70.965.375.491.850.686.250.5164K$0.259$0.800
38GLM 5.1(on/off开/关)Z.ai70.166.266.978.567.486.261205K$0.966$3.04
39Mistral Medium 3.5Mistral AI65.442.969.487.886.395.151262K$1.50$7.50
40Seed-2.0-LiteByteDance Seed65.366.972.9—50.695.132.2262K$0.250$2.00
41gpt-oss-120bOpenAI64.95956.385.393.466.435.5131K$0.150$0.600
42Gemini 3.5 Flash Lite(high)Google63.460.457.892.736—58.81M$0.300$2.50
43Gemini 3.1 Flash Lite(high)Google58.660.54989.634.3—52.21M$0.250$1.50
44Mercury 2Inception57.2 low data数据少53.828.8—91.1——128K$0.250$0.750
45Claude Haiku 4.5(high)Anthropic54.255.866.270.221.839.947.5200K$1.00$5.00
46Sonar ProPerplexity52.5 low data数据少44.3—81.529——200K$3.00$15
47Command ACohere50.736.531.476.99085.832.8256K$2.50$10
48Mistral Small 4Mistral AI50.231.458.757.19341.6—262K$0.150$0.600
49Mistral LargeMistral AI49.926.652.87963.4—48128K$2.00$6.00
50Command R+ (08-2024)Cohere42.8 low data数据少33—81.76.7——128K$2.50$10
51Llama 4 MaverickMeta42.541.539.686.9111848.51M$0.188$0.652
52Nova Pro 1.0Amazon39.1 low data数据少30—74.67——300K$0.800$3.20
53Phi 4Microsoft38.5 low data数据少32.8—77.16.7—8.316K$0.070$0.140
54Llama 4 ScoutMeta34.533.723.580.23.615.641.71.3M$0.180$0.590
55Nova Premier 1.0Amazon30.6 low data数据少33.433.9—17——1M$2.50$13
—GPT-5.6 Terra Pro(max)OpenAI— low data数据少72.877———60.21.1M$2.00$12
—LongCat 2.0Meituan— low data数据少81.7——100——1M$0.300$1.20
—Hunyuan A13B InstructTencent— low data数据少94.7——76.8——131K$0.140$0.570
—Solar Pro 3Upstage— low data数据少75.8——84.3——131K$0.150$0.600
—Palmyra X5Writer— low data数据少 vendor data厂商数据49.4—————1M$0.600$6.00

55 of 60 models have enough data (≥30% benchmark coverage) for a public score. Prices are API list prices per 1M tokens from OpenRouter. "vendor data" means >40% of available scores are vendor self-reported — treat with care. The Chinese board lists models reachable via CN-direct endpoints, ranked by score.55/60 个模型数据充足(benchmark 覆盖 ≥30%)予以出分。价格为 OpenRouter 每百万 token API 标价。"厂商数据" 表示可用成绩中厂商自报占比超 40%,请谨慎参考。中文榜 = 支持国内直连的模型按分数排序。