Benchmarks say one thing, people using the models say another. We pull both — plus real market pricing and measured speed — and score 207 models on one scale. 113 of them carry a machine score and a human score, which is where the disagreements show up.
24 of these models have a hands-on review on this site.
What feeds the index
live market pricing
benchmark indices
human preference votes
113 models scored by machines and humans
Benchmark indices cross-checked against human preference votes. The bar under each score splits it into the five things we weigh — same total, different shape means a different model for a different job.
Benchmarks, human votes, speed and real pricing on the same screen. The highlighted side wins that line.
Claude Opus 5
anthropic
90.0
intelligence · wins 11 lines
Claude Fable 5
anthropic
86.7
intelligence · wins 2 lines
Our composite score
| 63.3 | Overall | 59.4 |
|---|---|---|
| 90 | Intelligence40% of the total | 87 |
| 15 | Price / perf20% of the total | 1 |
| 85 | Capabilities15% of the total | 85 |
| 34 | Speed10% of the total | 37 |
| 55 | Ecosystem15% of the total | 55 |
Measured intelligence
| 60.7 | Intelligence indexArtificial Analysis | 59.9 |
|---|---|---|
| 78.0 | Coding indexArtificial Analysis | 76.5 |
| 55.3 | Agentic indexArtificial Analysis | 52.8 |
| — | Math indexArtificial Analysis | — |
| 1512 · 2k votes | Human preferenceLMArena Elo | 1494 · 16k votes |
Speed
| 54 t/s | Output throughputtokens/second | 61 t/s |
|---|---|---|
| 34.11 s | Time to first token | 47.51 s |
Price per million tokens
| $5.0 | Input | $10 |
|---|---|---|
| $25 | Output | $50 |
| $10 | Blended 3:1typical real-world mix | $20 |
Capabilities
| 1M | Context window | 1M |
|---|---|---|
| 128k | Max output | 128k |
| Yes | Reasoning | Yes |
| No | Open weights | No |
| text, image, file | Modalities | text, image, file |
| 1 | Providers serving it | 1 |
| — | Knowledge cutoff | — |
Filter it down to the models that fit your constraints, then open a row for the full measurement set.
207 of 207 models
| # | Model | intelligence & profile | Intel. | Elo | t/s | In / Out | Context | Actions |
|---|---|---|---|---|---|---|---|---|
| 1 | anthropic | 90.0 | 60.7 | 1512 | 54 | $5.0 / $25 | 1M | |
| 2 | anthropic | 86.7 | 59.9 | 1494 | 61 | $10 / $50 | 1M | |
| 3 | openai | 82.9 | 58.9 | — | 73 | $5.0 / $30 | 1.1M | |
| 4 | Kimi K3OPEN moonshotai | 82.5 | 57.1 | 1473 | 33 | $3.0 / $15 | 1.0M | |
| 5 | anthropic | 82.0 | 53.5 | 1490 | — | $5.0 / $25 | 1M | |
| 6 | openai | 80.0 | 54.8 | 1469 | — | $5.0 / $30 | 1.1M | |
| 7 | anthropic | 79.8 | 55.7 | 1463 | — | $5.0 / $25 | 1M | |
| 8 | meta | 78.0 | 50.6 | 1479 | 133 | $1.3 / $4.3 | 1.0M | |
| 9 | Qwen3.6 Max Preview qwen | 78.0 | — | 1446 | — | $1.0 / $6.2 | 262k | |
| 10 | 77.8 | 50.2 | 1480 | 221 | $1.5 / $9.0 | 1.0M | ||
| 11 | openai | 77.7 | 51.4 | 1470 | — | $2.5 / $15 | 1.1M | |
| 12 | openai | 77.5 | 55.0 | — | 123 | $1.0 / $6.0 | 1.1M | |
| 13 | Gemini 3.6 Flash google | 77.5 | 50.1 | 1478 | 195 | $1.5 / $7.5 | 1.0M | |
| 14 | x-ai | 77.1 | 53.8 | 1452 | 53 | $2.0 / $6.0 | 500k | |
| 15 | GLM 5.2OPEN z-ai | 76.9 | 51.1 | 1464 | 142 | $0.31 / $0.98 | 1.0M | |
| 16 | anthropic | 75.6 | 53.4 | 1440 | 75 | $2.0 / $10 | 1M | |
| 17 | Qwen3.7 Max qwen | 73.8 | 46.0 | 1475 | 198 | $1.5 / $4.4 | 1M | |
| 18 | DeepSeek V4 FlashOPEN deepseek | 73.6 | 49.9 | 1432 | — | $0.14 / $0.28 | 1.0M | |
| 19 | anthropic | 73.4 | 37.8 | 1503 | — | $5.0 / $25 | 1M | |
| 20 | 73.3 | 46.5 | 1480 | 120 | $2.0 / $12 | 1.0M | ||
| 21 | openai | 72.7 | 51.2 | — | 162 | $0.10 / $0.60 | 1.1M | |
| 22 | DeepSeek V4 Flash 0731OPEN deepseek | 72.6 | 49.9 | 1432 | — | $0.09 / $0.18 | 1.0M | |
| 23 | DeepSeek V4 ProOPEN deepseek | 70.1 | 44.3 | 1450 | 59 | $0.43 / $0.87 | 1.0M | |
| 24 | Kimi K2.6OPEN moonshotai | 69.9 | 44.2 | 1455 | — | $0.60 / $3.4 | 262k | |
| 25 | MiMo-V2.5-ProOPEN xiaomi | 69.7 | 42.2 | 1462 | 37 | $0.43 / $0.87 | 1.1M | |
| 26 | anthropic | 69.3 | 35.9 | 1457 | — | $3.0 / $15 | 1M | |
| 27 | Hy3OPEN tencent | 69.0 | 41.2 | 1442 | 70 | $0.13 / $0.53 | 262k | |
| 28 | GLM 5.1OPEN z-ai | 68.9 | 40.2 | 1465 | — | $0.97 / $3.0 | 205k | |
| 29 | Gemini 3 Flash Preview google | 68.6 | 37.8 | 1466 | — | $0.50 / $3.0 | 1.0M | |
| 30 | MiniMax M3OPEN minimax | 67.9 | 44.4 | 1436 | 89 | $0.30 / $1.2 | 1.0M | |
| 31 | GLM 5OPEN z-ai | 67.1 | 39.5 | 1445 | — | $0.95 / $2.5 | 205k | |
| 32 | Hy3 previewOPEN tencent | 66.8 | 41.2 | 1442 | 70 | $0.06 / $0.21 | 262k | |
| 33 | InklingOPEN thinkingmachines | 66.0 | 40.7 | 1442 | 77 | $1.0 / $4.0 | 1.0M | |
| 34 | Qwen3.7 Plus qwen | 65.7 | 39.0 | 1457 | 52 | $0.32 / $1.3 | 1M | |
| 35 | Qwen3.5-Flash qwen | 65.2 | — | 1398 | — | $0.07 / $0.26 | 1M | |
| 36 | GPT-5.2 openai | 65.2 | 42.2 | 1416 | — | $1.8 / $14 | 400k | |
| 37 | GPT-5.2 Chat openai | 65.2 | 42.2 | 1416 | — | $1.8 / $14 | 128k | |
| 38 | Qwen3.6 Plus qwen | 64.4 | 39.6 | 1437 | 55 | $0.33 / $1.9 | 1M | |
| 39 | GPT-5.3-Codex openai | 63.3 | 44.3 | — | 120 | $1.8 / $14 | 400k | |
| 40 | Gemini 3.5 Flash Lite google | 62.6 | 36.5 | 1439 | 342 | $0.30 / $2.5 | 1.0M | |
| 41 | GPT-5.3 Chat openai | 62.5 | — | 1388 | — | $1.8 / $14 | 128k | |
| 42 | GLM 5V Turbo z-ai | 62.4 | 34.5 | 1437 | — | $1.2 / $4.0 | 203k | |
| 43 | GPT-5.1 openai | 62.1 | 36.9 | 1441 | — | $1.3 / $10 | 400k | |
| 44 | GPT-5.4 Mini openai | 62.0 | 40.0 | 1413 | — | $0.75 / $4.5 | 400k | |
| 45 | MiMo-V2.5OPEN xiaomi | 61.9 | 37.2 | 1428 | — | $0.14 / $0.28 | 1.1M | |
| 46 | Kimi K2.5OPEN moonshotai | 61.9 | 35.4 | 1445 | — | $0.57 / $2.9 | 262k | |
| 47 | Qwen3 Max Thinking qwen | 60.7 | 31.7 | 1439 | — | $0.78 / $3.9 | 262k | |
| 48 | GLM 4.7OPEN z-ai | 60.5 | 33.7 | 1436 | — | $0.40 / $1.8 | 205k | |
| 49 | Qwen3.5 397B A17BOPEN qwen | 60.3 | 33.7 | 1439 | 63 | $0.39 / $2.3 | 262k | |
| 50 | DeepSeek V3.2 ExpOPEN deepseek | 59.5 | 32.0 | 1424 | — | $0.27 / $0.41 | 164k | |
| 51 | MiniMax M2.7OPEN minimax | 59.1 | 38.1 | 1405 | — | $0.25 / $1.0 | 205k | |
| 52 | Gemini 2.5 Pro Preview 06-05 google | 58.9 | 25.8 | 1458 | 131 | $1.3 / $10 | 1.0M | |
| 53 | Claude Opus 4.5 anthropic | 58.3 | 40.8 | — | — | $5.0 / $25 | 200k | |
| 54 | Qwen3.5-27BOPEN qwen | 58.1 | 33.8 | 1408 | — | $0.20 / $1.6 | 262k | |
| 55 | GLM 4.6OPEN z-ai | 58.0 | 28.7 | 1440 | — | $0.50 / $2.0 | 205k | |
| 56 | Gemma 4 31BOPEN google | 57.5 | 29.4 | 1442 | 35 | $0.10 / $0.34 | 262k | |
| 57 | Kimi K2.7 CodeOPEN moonshotai | 57.4 | 41.9 | — | 39 | $0.73 / $3.5 | 262k | |
| 58 | GPT-5.2-Codex openai | 57.3 | 40.1 | — | — | $1.8 / $14 | 400k | |
| 59 | DeepSeek V3.2OPEN deepseek | 56.9 | 32.0 | 1424 | — | $0.27 / $0.40 | 164k | |
| 60 | Nex-N2-ProOPEN nex-agi | 56.8 | 41.0 | — | 131 | $0.25 / $1.0 | 262k | |
| 61 | Qwen3.5-122B-A10BOPEN qwen | 56.8 | 32.3 | 1418 | 138 | $0.26 / $2.1 | 262k | |
| 62 | x-ai | 56.5 | 37.6 | 1400 | — | $1.3 / $2.5 | 1M | |
| 63 | Mistral Medium 3.5 mistralai | 56.3 | 29.9 | 1421 | 81 | $1.5 / $7.5 | 262k | |
| 64 | openai | 55.9 | 34.7 | 1406 | — | $1.3 / $10 | 400k | |
| 65 | GPT-5.4 Nano openai | 55.7 | 38.2 | 1373 | — | $0.20 / $1.3 | 400k | |
| 66 | DeepSeek V3.1 TerminusOPEN deepseek | 55.6 | 30.4 | 1420 | — | $0.27 / $1.0 | 164k | |
| 67 | Gemini 2.5 Pro google | 55.6 | 25.8 | 1458 | 131 | $1.3 / $10 | 1.0M | |
| 68 | Qwen3 Max qwen | 55.2 | 24.0 | 1439 | — | $0.78 / $3.9 | 262k | |
| 69 | Inkling SmallOPEN thinkingmachines | 54.4 | 40.2 | — | 113 | $0.50 / $1.2 | 524k | |
| 70 | GLM 5 Turbo z-ai | 54.4 | 38.1 | — | — | $1.2 / $4.0 | 203k | |
| 71 | MiniMax M2.1OPEN minimax | 54.2 | 31.4 | 1391 | — | $0.30 / $1.2 | 205k | |
| 72 | Gemma 4 26B A4B OPEN google | 54.1 | 25.7 | 1435 | — | $0.07 / $0.34 | 262k | |
| 73 | Gemini 3.1 Flash Lite google | 53.2 | 25.0 | 1415 | 276 | $0.25 / $1.5 | 1.0M | |
| 74 | x-ai | 52.9 | 37.0 | — | — | $1.3 / $2.5 | 2M | |
| 75 | Grok Build 0.1 x-ai | 52.7 | 39.8 | — | — | $1.0 / $2.0 | 256k | |
| 76 | Kimi K2 0711OPEN moonshotai | 52.4 | 32.7 | 1371 | — | $0.57 / $2.3 | 131k | |
| 77 | Step 3.5 FlashOPEN stepfun | 52.0 | 26.0 | 1404 | — | $0.10 / $0.30 | 262k | |
| 78 | MiniMax M2.5OPEN minimax | 51.5 | 33.7 | 1359 | — | $0.15 / $0.90 | 205k | |
| 79 | Qwen3.6 27BOPEN qwen | 51.1 | 37.1 | — | 56 | $0.30 / $2.0 | 262k | |
| 80 | R1 0528OPEN deepseek | 50.9 | 20.1 | 1428 | — | $0.50 / $2.1 | 164k | |
| 81 | MiniMax M1 minimax | 50.7 | — | 1343 | — | $0.55 / $2.2 | 1M | |
| 82 | GLM 4.5OPEN z-ai | 50.6 | 19.5 | 1429 | — | $0.60 / $2.2 | 131k | |
| 83 | Nemotron 3 Ultra (free)OPEN nvidia | 50.5 | 37.8 | — | — | Free / Free | 1M | |
| 84 | Nemotron 3 UltraOPEN nvidia | 50.5 | 37.8 | — | — | $0.60 / $3.6 | 512k | |
| 85 | DeepSeek V3.1OPEN deepseek | 50.4 | 21.0 | 1419 | — | $0.25 / $0.95 | 164k | |
| 86 | Qwen3 VL 235B A22B InstructOPEN qwen | 50.4 | 20.6 | 1422 | — | $0.21 / $1.9 | 262k | |
| 87 | Qwen3 VL 235B A22B ThinkingOPEN qwen | 50.4 | 20.6 | 1422 | — | $0.40 / $4.0 | 131k | |
| 88 | Qwen3.5-35B-A3BOPEN qwen | 50.0 | 29.3 | 1396 | — | $0.14 / $1.0 | 262k | |
| 89 | Gemini 3.1 Flash Lite Preview google | 49.7 | 25.0 | 1415 | 276 | $0.25 / $1.5 | 1.0M | |
| 90 | Gemini 2.5 Flash google | 49.6 | 20.1 | 1418 | — | $0.30 / $2.5 | 1.0M | |
| 91 | GPT-5.1-Codex openai | 49.6 | 34.7 | — | — | $1.3 / $10 | 400k | |
| 92 | GPT-5.1-Codex-Max openai | 49.6 | 34.7 | — | — | $1.3 / $10 | 400k | |
| 93 | Claude Sonnet 4.5 anthropic | 49.3 | 36.4 | — | — | $3.0 / $15 | 1M | |
| 94 | Qwen3 235B A22B Instruct 2507OPEN qwen | 48.9 | 19.6 | 1419 | — | $0.09 / $0.55 | 262k | |
| 95 | LongCat 2.0OPEN meituan | 48.7 | 33.5 | — | 43 | $0.30 / $1.2 | 1.0M | |
| 96 | Kimi K2 0905OPEN moonshotai | 47.0 | 23.5 | 1379 | — | $0.60 / $2.5 | 262k | |
| 97 | Qwen3 Next 80B A3B InstructOPEN qwen | 46.5 | 16.7 | 1419 | 202 | $0.10 / $1.1 | 262k | |
| 98 | o3 Pro openai | 46.4 | 32.5 | — | — | $20 / $80 | 200k | |
| 99 | KAT-Coder-Pro V2 kwaipilot | 46.1 | 33.7 | — | 105 | $0.30 / $1.2 | 262k | |
| 100 | GPT-5 Mini openai | 46.1 | 30.9 | 1374 | — | $0.25 / $2.0 | 400k | |
| 101 | MiniMax M2OPEN minimax | 45.5 | 28.3 | 1342 | — | $0.26 / $1.0 | 205k | |
| 102 | o1 openai | 45.2 | 23.4 | 1353 | — | $15 / $60 | 200k | |
| 103 | Qwen3 Next 80B A3B ThinkingOPEN qwen | 44.1 | 16.7 | 1419 | 202 | $0.15 / $1.2 | 262k | |
| 104 | GPT-5.1-Codex-Mini openai | 43.7 | 30.6 | — | — | $0.25 / $2.0 | 400k | |
| 105 | gpt-oss-120bOPEN openai | 43.5 | 23.8 | 1366 | 215 | $0.04 / $0.17 | 131k | |
| 106 | o3 openai | 43.4 | 30.4 | — | 135 | $2.0 / $8.0 | 200k | |
| 107 | Qwen3 235B A22B Thinking 2507OPEN qwen | 43.2 | 13.4 | 1414 | — | $0.23 / $2.3 | 262k | |
| 108 | GLM 4.7 FlashOPEN z-ai | 43.1 | 22.9 | 1353 | — | $0.06 / $0.40 | 203k | |
| 109 | GLM 4.5 AirOPEN z-ai | 42.4 | 16.5 | 1383 | — | $0.13 / $0.85 | 131k | |
| 110 | Qwen3.6 35B A3BOPEN qwen | 41.9 | 31.6 | — | 110 | $0.14 / $1.0 | 262k | |
| 111 | GLM 4.6VOPEN z-ai | 41.6 | 16.8 | 1375 | — | $0.30 / $0.90 | 131k | |
| 112 | Gemma 3n 4BOPEN google | 41.1 | — | 1306 | — | $0.06 / $0.12 | 33k | |
| 113 | Mercury 2 inception | 41.0 | 21.4 | 1358 | 822 | $0.25 / $0.75 | 128k | |
| 114 | Ring-2.6-1T inclusionai | 40.5 | 30.6 | — | 128 | $0.07 / $0.63 | 262k | |
| 115 | R1OPEN deepseek | 40.5 | 20.1 | 1373 | — | $0.70 / $2.5 | 164k | |
| 116 | Step 3.7 FlashOPEN stepfun | 40.3 | 30.3 | — | 392 | $0.20 / $1.1 | 262k | |
| 117 | Qwen3 Coder 480B A35BOPEN qwen | 40.0 | 18.0 | 1356 | — | $0.30 / $1.0 | 262k | |
| 118 | anthropic | 39.2 | 29.6 | — | — | $1.0 / $5.0 | 200k | |
| 119 | DeepSeek V3 0324OPEN deepseek | 38.3 | 15.4 | 1375 | — | $0.27 / $1.1 | 164k | |
| 120 | Gemma 4 31B (free)OPEN google | 38.2 | 29.4 | — | — | Free / Free | 262k | |
| 121 | Nova 2 Lite amazon | 38.2 | 18.2 | 1363 | — | $0.30 / $2.5 | 1M | |
| 122 | o3 Mini openai | 38.1 | 19.0 | 1337 | — | $1.1 / $4.4 | 200k | |
| 123 | Qwen3 235B A22BOPEN qwen | 38.0 | 13.4 | 1366 | — | $0.46 / $1.8 | 131k | |
| 124 | Qwen3 30B A3B Instruct 2507OPEN qwen | 37.4 | 9.3 | 1384 | — | $0.05 / $0.19 | 262k | |
| 125 | Ling-2.6-1T inclusionai | 37.3 | 26.1 | — | — | $0.07 / $0.63 | 262k | |
| 126 | Claude Sonnet 4 anthropic | 37.0 | 28.9 | — | — | $3.0 / $15 | 1M | |
| 127 | GPT-5 Nano openai | 36.6 | 19.9 | 1320 | — | $0.05 / $0.40 | 400k | |
| 128 | o4 Mini High openai | 36.6 | 25.6 | — | — | $1.1 / $4.4 | 200k | |
| 129 | o4 Mini openai | 36.6 | 25.6 | — | — | $1.1 / $4.4 | 200k | |
| 130 | Trinity Large ThinkingOPEN arcee-ai | 36.0 | 18.2 | 1342 | 201 | $0.22 / $0.85 | 262k | |
| 131 | DeepSeek V3OPEN deepseek | 35.0 | 14.2 | 1333 | — | $0.26 / $1.0 | 164k | |
| 132 | o3 Mini High openai | 33.9 | 19.0 | 1337 | — | $1.1 / $4.4 | 200k | |
| 133 | Gemma 4 26B A4B (free)OPEN google | 33.2 | 25.7 | — | — | Free / Free | 262k | |
| 134 | Nemotron 3 SuperOPEN nvidia | 31.7 | 25.4 | — | — | $0.09 / $0.40 | 1M | |
| 135 | Nemotron 3 Super (free)OPEN nvidia | 31.7 | 25.4 | — | — | Free / Free | 262k | |
| 136 | Qwen3 32BOPEN qwen | 31.6 | 11.5 | 1340 | — | $0.08 / $0.28 | 131k | |
| 137 | Gemma 3 27BOPEN google | 31.5 | 7.4 | 1358 | — | $0.08 / $0.45 | 262k | |
| 138 | GLM 4.5VOPEN z-ai | 30.7 | 9.1 | 1334 | — | $0.60 / $1.8 | 66k | |
| 139 | Kimi K2 ThinkingOPEN moonshotai | 29.8 | 32.7 | — | — | $0.60 / $2.5 | 262k | |
| 140 | Llama 3.1 70B InstructOPEN meta-llama | 29.2 | — | 1261 | — | $0.40 / $0.40 | 131k | |
| 141 | Qwen3 30B A3BOPEN qwen | 28.6 | 9.3 | 1317 | — | $0.12 / $0.50 | 131k | |
| 142 | GPT-4o (2024-05-13) openai | 28.4 | 8.6 | 1301 | — | $5.0 / $15 | 128k | |
| 143 | Qwen3 Coder NextOPEN qwen | 28.3 | 21.1 | — | 126 | $0.12 / $0.80 | 262k | |
| 144 | GPT-4.1 openai | 27.7 | 19.4 | — | — | $2.0 / $8.0 | 1.0M | |
| 145 | Qwen3 30B A3B Thinking 2507OPEN qwen | 27.3 | 9.3 | 1317 | — | $0.20 / $2.4 | 82k | |
| 146 | Gemma 3 12BOPEN google | 27.1 | 5.5 | 1334 | — | $0.05 / $0.15 | 131k | |
| 147 | o1-pro openai | 27.0 | 18.9 | — | — | $150 / $600 | 200k | |
| 148 | gpt-oss-20bOPEN openai | 26.9 | 14.9 | 1288 | 190 | $0.03 / $0.13 | 131k | |
| 149 | Qwen3.5-9BOPEN qwen | 25.9 | 21.4 | — | 65 | $0.10 / $0.15 | 262k | |
| 150 | Qwen3 VL 32B InstructOPEN qwen | 25.6 | 17.9 | — | — | $0.10 / $0.42 | 131k | |
| 151 | North Mini Code (free)OPEN cohere | 25.4 | 19.8 | — | — | Free / Free | 256k | |
| 152 | Sonar Reasoning Pro perplexity | 25.4 | 17.8 | — | — | $2.0 / $8.0 | 128k | |
| 153 | GPT-4o (2024-08-06) openai | 24.3 | 9.6 | 1283 | — | $2.5 / $10 | 128k | |
| 154 | Olmo 3 32B ThinkOPEN allenai | 24.2 | 6.4 | 1299 | — | $0.15 / $0.50 | 66k | |
| 155 | Granite 4.1 8BOPEN ibm-granite | 23.7 | 6.7 | 1293 | 121 | $0.05 / $0.10 | 131k | |
| 156 | GPT-4o-mini (2024-07-18) openai | 23.4 | 6.9 | 1287 | — | $0.15 / $0.60 | 128k | |
| 157 | Mistral Small 4OPEN mistralai | 23.1 | 19.6 | — | 151 | $0.15 / $0.60 | 262k | |
| 158 | Llama 3.3 70B InstructOPEN meta-llama | 21.5 | 9.4 | 1275 | — | $0.13 / $0.40 | 131k | |
| 159 | Gemma 2 27BOPEN google | 21.4 | — | 1232 | — | $0.65 / $0.65 | 8k | |
| 160 | Command R+ (08-2024) cohere | 20.8 | — | 1229 | — | $2.5 / $10 | 128k | |
| 161 | Mistral Large 2407 mistralai | 20.5 | 7.3 | 1266 | — | $2.0 / $6.0 | 131k | |
| 162 | Gemma 3 4BOPEN google | 19.6 | 1.1 | 1291 | — | $0.05 / $0.10 | 131k | |
| 163 | Qwen3 Coder 30B A3B InstructOPEN qwen | 19.4 | 13.6 | — | — | $0.07 / $0.28 | 262k | |
| 164 | Qwen3 VL 30B A3B InstructOPEN qwen | 19.0 | 13.3 | — | — | $0.13 / $0.52 | 262k | |
| 165 | Qwen3 VL 30B A3B ThinkingOPEN qwen | 19.0 | 13.3 | — | — | $0.20 / $2.4 | 262k | |
| 166 | Mistral Medium 3.1 mistralai | 18.4 | 14.7 | — | — | $0.40 / $2.0 | 131k | |
| 167 | Ling-2.6-flash inclusionai | 17.9 | 14.1 | — | 85 | $0.01 / $0.03 | 262k | |
| 168 | Mistral Medium 3 mistralai | 17.9 | 12.5 | — | — | $0.40 / $2.0 | 131k | |
| 169 | gpt-oss-20b (free)OPEN openai | 17.5 | 14.9 | — | — | Free / Free | 131k | |
| 170 | GPT-4.1 Mini openai | 16.8 | 14.8 | — | — | $0.40 / $1.6 | 1.0M | |
| 171 | Sonar perplexity | 16.7 | 11.7 | — | — | $1.0 / $1.0 | 127k | |
| 172 | Command AOPEN cohere | 16.5 | 7.7 | — | 62 | $2.5 / $10 | 256k | |
| 173 | Gemini 2.5 Flash Lite google | 16.3 | 11.4 | — | — | $0.10 / $0.40 | 1.0M | |
| 174 | GPT-4o openai | 16.0 | 11.2 | — | — | $2.5 / $10 | 128k | |
| 175 | GPT-4o (2024-11-20) openai | 16.0 | 11.2 | — | — | $2.5 / $10 | 128k | |
| 176 | Mistral Small 3OPEN mistralai | 15.9 | 6.9 | 1234 | — | $0.05 / $0.08 | 33k | |
| 177 | Solar Pro 3 upstage | 15.5 | 14.1 | — | 15 | $0.15 / $0.60 | 128k | |
| 178 | GPT-4 Turbo openai | 15.5 | 7.9 | — | — | $10 / $30 | 128k | |
| 179 | GPT-4 Turbo Preview openai | 15.5 | 7.9 | — | — | $10 / $30 | 128k | |
| 180 | Llama 4 MaverickOPEN meta-llama | 15.2 | 14.3 | — | 106 | $0.20 / $0.80 | 1.0M | |
| 181 | Qwen3 VL 8B InstructOPEN qwen | 15.1 | 10.6 | — | — | $0.12 / $0.46 | 262k | |
| 182 | Qwen3 VL 8B ThinkingOPEN qwen | 15.1 | 10.6 | — | — | $0.18 / $2.1 | 131k | |
| 183 | Nemotron 3 Nano 30B A3BOPEN nvidia | 14.9 | 14.2 | — | — | $0.05 / $0.20 | 262k | |
| 184 | Nemotron 3 Nano 30B A3B (free)OPEN nvidia | 14.9 | 14.2 | — | — | Free / Free | 256k | |
| 185 | R1 Distill Llama 70BOPEN deepseek | 14.1 | 9.9 | — | — | $0.80 / $0.80 | 8k | |
| 186 | Sonar Pro perplexity | 13.3 | 9.3 | — | — | $3.0 / $15 | 200k | |
| 187 | Ministral 3 14B 2512OPEN mistralai | 12.7 | 11.1 | — | — | $0.20 / $0.20 | 262k | |
| 188 | Phi 4OPEN microsoft | 12.3 | 4.9 | 1217 | — | $0.07 / $0.14 | 16k | |
| 189 | Qwen3 14BOPEN qwen | 11.9 | 10.4 | — | — | $0.23 / $0.91 | 131k | |
| 190 | GPT-4 openai | 11.5 | 7.0 | — | — | $30 / $60 | 8k | |
| 191 | Mistral Large 3 2512 mistralai | 10.7 | 4.4 | — | — | $0.50 / $1.5 | 262k | |
| 192 | GPT-4.1 Nano openai | 10.4 | 9.6 | — | — | $0.10 / $0.40 | 1.0M | |
| 193 | Command R (08-2024) cohere | 9.9 | — | 1188 | — | $0.15 / $0.60 | 128k | |
| 194 | Llama 4 ScoutOPEN meta-llama | 9.8 | 10.0 | — | 112 | $0.10 / $0.30 | 1.3M | |
| 195 | Ministral 3 8B 2512OPEN mistralai | 9.6 | 9.0 | — | — | $0.15 / $0.15 | 262k | |
| 196 | Saba mistralai | 9.1 | 6.4 | — | — | $0.20 / $0.60 | 33k | |
| 197 | Qwen3 8BOPEN qwen | 9.0 | 8.3 | — | — | $0.12 / $0.46 | 131k | |
| 198 | GPT-4o-mini openai | 8.5 | 6.9 | — | — | $0.15 / $0.60 | 128k | |
| 199 | Llama 3.1 8B InstructOPEN meta-llama | 8.4 | 7.6 | 1187 | — | $0.05 / $0.08 | 131k | |
| 200 | Hermes 3 70B InstructOPEN nousresearch | 7.3 | 5.1 | — | — | $0.70 / $0.70 | 131k | |
| 201 | Ministral 3 3B 2512OPEN mistralai | 6.8 | 6.8 | — | — | $0.10 / $0.10 | 131k | |
| 202 | Mistral Large mistralai | 6.3 | 4.4 | — | — | $2.0 / $6.0 | 128k | |
| 203 | Reka Flash 3OPEN rekaai | 5.9 | 4.1 | — | 98 | $0.10 / $0.20 | 66k | |
| 204 | Claude 3 Haiku anthropic | 5.6 | 3.9 | — | — | $0.25 / $1.3 | 200k | |
| 205 | Granite 4.0 MicroOPEN ibm-granite | 3.4 | 2.4 | — | — | $0.02 / $0.11 | 131k | |
| 206 | Llama 3.2 3B InstructOPEN meta-llama | 0.0 | — | 1110 | — | $0.05 / $0.33 | 131k | |
| 207 | Llama 3.2 1B InstructOPEN meta-llama | 0.0 | — | 1055 | — | $0.03 / $0.20 | 60k |
The raw measurements come from third parties, credited below. What we add is the cross-source aggregation and the scoring: five axes on absolute 0-100 scales, weighted into one number. Absolute scales matter — a model scoring 78 today means the same as a model scoring 78 six months ago, so the history stays comparable as new models arrive.
Average of every published capability index we can cross-check — general intelligence, coding and agentic benchmarks.
Real market price per million tokens weighted against measured capability. A cheap weak model and an expensive strong one can score the same.
Context window, maximum output, modalities, reasoning support, prompt caching and how recent the knowledge cutoff is.
Measured output throughput (tokens per second) and time to first token. A model you wait ten seconds for is a different product from one that answers instantly.
How many providers serve the model, and whether the weights are open — availability and independence from a single vendor.
Every collection is archived, so price cuts and ranking moves stay traceable. The sources publish today's numbers; this is the trail behind them.
The default ranking is capability, so the most expensive model can sit above a free one. Switch the criterion at the top of the page and the order changes completely — that is the point. Four questions come up constantly, and here are the answers.
Because the default ranking measures what a model can do, not what it costs. Price is a separate axis, and it has its own ranking one click away. Ranking on the composite score instead buried the frontier models behind cheap ones and turned the whole index into a price chart, which is not what a ranked list of models is for. Today that puts Claude Opus 5 first with 90.0 on capability at $25.00 per million output tokens.
Both count, equally. Benchmarks test verifiable problems; the arena records which answer people actually preferred. When a model carries both, each contributes half its intelligence score. The widest gap in the index right now is Qwen: Qwen3 235B A22B Thinking 2507: 19 from benchmarks against 69 from human votes — a 50-point disagreement that a single-source ranking would hide entirely.
The score itself, split into the five things we weigh. Each segment is that axis's contribution, so the full bar length is the composite score. Two models can share a total and have opposite shapes — one fast and cheap, the other slow and brilliant.
Because that measurement does not exist yet, and we would rather show a dash than invent a number. Of the 207 models listed, 196 carry benchmark indices and 124 carry human ratings. A missing value never counts as zero — it simply leaves that axis out of the model's score, and the “Double-validated” filter isolates the 113 models measured both ways.
A ranking of 207 large language models built by cross-referencing three independent sources: live market pricing from OpenRouter, benchmark indices from Artificial Analysis, and human preference votes from LMArena. We aggregate those measurements and compute our own score across five weighted axes. The raw measurements belong to their sources; the scoring and the ranking are ours.
By measured capability, not by price. A model that costs ten times more but answers harder questions ranks higher by default. Three alternative orderings are one click away — price/performance, speed, and the weighted composite score — because "best" means something different depending on whether you are paying the bill or waiting for the response.
Two things, weighted equally when both exist. Machine benchmarks contribute the general intelligence index (50%), the coding index (25%) and the agentic index (25%), each normalised against its own ceiling. Human preference contributes the LMArena Elo rating, mapped onto the same 0-100 scale. A model measured by only one of the two is scored on that one alone.
A model carrying both a machine benchmark score and a human preference rating. 113 of the 207 models in the index qualify. It matters because the two methods disagree more often than you would expect: benchmarks measure what is testable, the arena measures what people actually prefer after using the model.
They measure different things. Benchmarks test verifiable problems with known answers — maths, code that compiles, factual recall. Human votes reward tone, instruction-following, formatting and refusal behaviour, which no benchmark captures. A model can top the benchmarks and lose blind head-to-head votes, and the reverse happens just as often.
Because nearly every current model scores in the 90s on it, so it separates nothing. Worse, averaging it in rewarded models that had been measured on maths and almost nothing else — one model with no coding or agentic score reached a higher intelligence score than models measured on everything. The maths index is still displayed as a raw measurement on each model.
Measured capability divided by the blended price per million tokens, weighted one part input to three parts output, which reflects typical real-world usage. The result is placed on a logarithmic scale because prices span four orders of magnitude. Free or unpriced models receive a neutral 50 rather than a perfect score, since they are not comparable.
Yes. They come from OpenRouter, which publishes live per-token rates charged by the providers actually serving each model, in US dollars per million tokens. Input and output are shown separately because the gap between them is often large and matters more than the headline figure.
Every 48 hours. Each collection re-fetches all three sources, recalculates every score, and writes a historical snapshot so price changes and ranking movements stay traceable. Models that disappear from the sources are marked inactive rather than deleted, so their history survives.
A model whose weights can be downloaded and self-hosted, as opposed to one reachable only through a vendor API. 111 of the 207 models qualify. Open weights add points on the ecosystem axis because they remove vendor lock-in, but a closed model served by many providers is not penalised to zero.
So the numbers stay comparable over time. If scores were normalised against the current field, every new model release would silently move every existing score, and a snapshot from six months ago would mean nothing. On an absolute scale, a model scoring 78 today means the same as a model scoring 78 last year.
24 of the 207 models link to a full review on this site, written after actually using the model. Those links appear directly on each model card and in the comparison table. The rest are scored from published measurements only, which is stated rather than hidden.
All raw metrics belong to their respective sources and are reproduced with attribution. The composite scores, weighting and rankings on this page are ThePlanetTools.ai's own computation and should be attributed to us, not to the data providers.
We're developers and SaaS builders who use these tools daily in production. Every review comes from hands-on experience building real products — DealPropFirm, ThePlanetIndicator, PropFirmsCodes, and many more. We don't just review tools — we build and ship with them every day.
Written and tested by developers who build with these tools daily.