Closed models
Closed models are available only through the developer's own products or API. Each model appears once, at its best-scoring setting. Where two models tie, they keep the order of the Artificial Analysis table.
| Rank | Model | Developer | Setting | Index |
|---|---|---|---|---|
| 01 | Claude Opus 5.5 | Anthropic | max | 58 |
| 02 | Claude Sonnet 5.5 | Anthropic | max | 56 |
| 03 | Muse Wonder Full | Meta | max | 56 |
| 04 | Claude Fable 5.1 | Anthropic | max | 53 |
| 05 | GPT-6 Astra | OpenAI | max | 53 |
| 06 | Gemini 4 Argon | high | 53 | |
| 07 | Muse Wonder | Meta | max | 52 |
| 08 | GPT-6.1 Sol | OpenAI | max | 52 |
| 09 | Claude Opus 5 * | Anthropic | aggregators | 50.8 |
| 10 | Claude Fable 5 * | Anthropic | one aggregator | 49.6 |
* Ranks 9 and 10 are not on the Artificial Analysis pages that could be read in full. They use the aggregator sites BenchLM and modelgrep, carry decimal scores that are not directly comparable, and rank 10 rests on a single aggregator.
Open-weight models
Open-weight models publish their trained weights so anyone can download and run them. That is not the same as open source: licences range from permissive (MIT) to custom community terms, so check the licence before building on one.
| Rank | Model | Developer | Licence | Index |
|---|---|---|---|---|
| 01 | MiMo-V2.6-Pro | Xiaomi | MIT | 46 |
| 02 | DeepSeek V4.1 Flash 1017 | DeepSeek | MIT | 45 |
| 03 | GLM-5.3 | Z.AI | GLM-5.3 License (custom) | 45 |
| 04 | Step 5 Preview | StepFun | StepFun Model License 1.0 (custom) | 44 |
| 05 | GLM-5.3-Flash | Z.AI | Not verified | 42 |
| 06 | Muse Spark Open | Meta | Muse Model License (custom) | 39 |
| 07 | Qwen3.8 27B | Alibaba | Not verified | 34 |
| 08 | K2 Horizon 375B A23B | Institute of Foundation Models | Not verified | 31 |
| 09 | Beam 1.0 | Reflection | Not verified | 30 |
| 10 | Nemotron 3.5 Super VL | NVIDIA | Not verified | 26 |
Licences are shown only where BenchLM's open-source page stated them. "Not verified" means no source read for this edition gave the licence; it does not mean the model has none.
Rank for rank, closed vs open
Closed ranks 9 and 10 use aggregator scores, so the right-hand pairs are less exact than the rest.
Developers in this edition
Closed top ten
- Anthropic5
- Meta2
- OpenAI2
- Google1
Open-weight top ten
- Z.AI2
- Alibaba1
- DeepSeek1
- Institute of Foundation Models1
- Meta1
- NVIDIA1
- Reflection1
- StepFun1
- Xiaomi1
Counts of models per developer in this snapshot, including the two aggregator-sourced closed entries.
Watchlist: unclear weight status
Two models score high enough to appear in the top ten, but their weight status is unsettled in the sources, so they sit outside both lists until it is.
| Model | Developer | Index | Status in sources |
|---|---|---|---|
| Kimi K3.1 Pro | Moonshot AI | 55 | Weights release listed as pending |
| Qwen3.8 Max | Alibaba | 45 | Flagged as not open weights |
Method and caveats
One yardstick. Scores are the Artificial Analysis Intelligence Index, a composite of benchmarks. It is a useful single number, not a measure of every task, and a model that scores lower can still be the better choice for a specific job.
Whole numbers mean ties. The Index is reported in whole numbers, so several models share a score. Their order here follows the source table and should not be read as a decisive gap.
Sources differ. Other aggregators report different scores for some of the same models, so this page uses Artificial Analysis wherever it could be read, and says so where it could not.
Settings matter. The same model can score several points apart at different reasoning settings. This page shows the best setting listed.
Editions and archive
Each edition is a dated snapshot. When a new one is published, this page moves to the archive and stays online unchanged.
| Edition | Status | Closed leader | Open leader |
|---|---|---|---|
| 25 October 2026 | Current | Claude Opus 5.5 (58) | MiMo-V2.6-Pro (46) |
Where the numbers come from
- Artificial Analysis — LLM leaderboard (closed ranks 1–8)
- Artificial Analysis — open-weights models (all open ranks)
- BenchLM — open-source LLM leaderboard (licences)
- BenchLM — Artificial Analysis Intelligence Index table (closed ranks 9–10)
- modelgrep — LLM leaderboard (closed ranks 9–10)
All figures read on 25 October 2026. Leaderboards update continuously, so a later visit to the sources may show different scores.
Buy me a coffee