Artificial Analysis Intelligence Index
Artificial Analysis runs a set of evaluations on each model and combines them with fixed weights into an Intelligence Index; version 4.3.2 uses ten evaluations. It also reports a Coding Index. The numbers shown here come through OpenRouter's benchmarks API, which lists a coding index and an agentic index next to the intelligence index. Because these are composites, a single number hides where a model is strong or weak, and the set of evaluations has changed between versions. Use it to place a model roughly, then look at the individual results.
Sources: Artificial Analysis intelligence benchmarking methodology
Intelligence Index against memory
Artificial Analysis · data from 2026-10-09
Not beaten on both memory and score● memory stated by the publisher◇ memory estimated
| # | Model | Configuration | Min memory | Intelligence Index | Frontier |
|---|---|---|---|---|---|
| 1 | Qwen3.8 27B | 27B Original weights | 54.3 GB ◇ | 33.7 | On frontier |
| 2 | North Mini Code 1.0 | Original weights | 58.1 GB ◇ | 9.9 | — |
| 3 | Nemotron 3.5 Lightning | 30B-A3B Original weights BF16 | 62.3 GB ◇ | 12.9 | — |
| 4 | Nemotron 3 Nano Omni | 30B-A3B Original weights BF16 | 62.5 GB ◇ | 10.3 | — |
| 5 | DeepSeek V4 Flash Vision Exp | Original weights | 158 GB ◇ | 34.8 | On frontier |
| 6 | Ling 3.0 Flash VL | Original weights | 238 GB ◇ | 24.6 | — |
| 7 | GLM 5.3 Flash | Original weights | 306 GB ◇ | 41.8 | On frontier |
| 8 | Step 3.7 Flash | Original weights | 387 GB ◇ | 19.5 | — |
| 9 | DeepSeek V4.1 Flash | Original weights | 476 GB ◇ | 39.5 | — |
| 10 | Kimi K2.7 Code | Original weights | 568 GB ◇ | 25.8 | — |
| 11 | GLM 5.3 | Original weights | 734 GB ◇ | 44.8 | On frontier |
| 12 | MiniMax M3 | Original weights | 797 GB ◇ | 29.2 | — |
| 13 | DeepSeek V4 Pro | Original weights | 807 GB ◇ | 36.0 | — |
| 14 | Nemotron 3 Ultra | 550B-A55B Original weights BF16 | 1045 GB ◇ | 22.9 | — |
| 15 | Kimi K3 | Original weights | 1475 GB ◇ | 43.6 | — |
| 16 | LongCat 2.0 | Original weights | 3308 GB ◇ | 19.1 | — |
| 17 | Qwen3.8 2.4T-A95B | 2.4T-A95B Original weights | 4560 GB ◇ | 39.9 | — |
Scores were measured on hosted endpoints, so each model is plotted once, at the memory need of its original weights. A quantized file needs less memory and may score differently. Lines connect the frontier points and do not predict results in between. Source: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings). artificialanalysis.ai
Open-weight models
| Model | Endpoint | Intelligence | Coding | Agentic | Original weights |
|---|---|---|---|---|---|
| GLM 5.3 | GLM-5.3 (Max) | 44.8 | 74.8 | 53.1 | 734 GB |
| Kimi K3 | Kimi K3 (Max) | 43.6 | 76.2 | 50.0 | 1475 GB |
| GLM 5.3 Flash | GLM-5.3-Flash | 41.8 | 71.5 | 50.9 | 306 GB |
| Qwen3.8 2.4T-A95B | Qwen3.8 2.4T A95B | 39.9 | 71.9 | 50.1 | 4560 GB |
| DeepSeek V4.1 Flash | DeepSeek V4.1 Flash (Max) | 39.5 | — | — | 476 GB |
| DeepSeek V4 Pro | DeepSeek V4 Pro 0813 (Max) | 36.0 | 68.8 | 41.3 | 807 GB |
| DeepSeek V4 Flash Vision Exp | DeepSeek V4 Flash Vision (Max) | 34.8 | 65.0 | 47.5 | 158 GB |
| Qwen3.8 27B | Qwen3.8 27B (Xhigh) | 33.7 | 68.1 | 45.8 | 54.3 GB |
| DeepSeek V4 Pro | DeepSeek V4 Pro 0424 (Max) | 30.4 | 59.4 | 26.3 | 807 GB |
| MiniMax M3 | MiniMax-M3 | 29.2 | 58.6 | 29.5 | 797 GB |
| Kimi K2.7 Code | Kimi K2.7 Code | 25.8 | 60.8 | 21.0 | 568 GB |
| Ling 3.0 Flash VL | Ling-3.0-flash-VL | 24.6 | 57.0 | 28.7 | 238 GB |
| Nemotron 3 Ultra | Nemotron 3 Ultra 550B A55B (Reasoning) | 22.9 | 49.3 | 20.1 | 1045 GB |
| Step 3.7 Flash | Step 3.7 Flash | 19.5 | 39.6 | — | 387 GB |
| LongCat 2.0 | LongCat 2.0 | 19.1 | 45.3 | 14.0 | 3308 GB |
| Nemotron 3.5 Lightning | Nemotron 3.5 Lightning | 12.9 | 26.8 | 3.5 | 62.3 GB |
| Nemotron 3 Nano Omni | Nemotron 3 Nano Omni 30B A3B Reasoning | 10.3 | 13.8 | — | 62.5 GB |
| North Mini Code 1.0 | North Mini Code | 9.9 | 36.5 | 1.1 | 58.1 GB |
Measured by Artificial Analysis on hosted endpoints, not by this site. A quantized file you run yourself may score differently. One model can have several endpoints. Source: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings). artificialanalysis.ai