AI model leaderboards
Frontier and open-weight model rankings, what developers actually run, and the most-downloaded open video models. Each board links to its source.
Frontier models
Blind head-to-head votes from real users, aggregated into an Elo rating. Higher means the model won more pairwise comparisons. Reasoning-effort variants are collapsed to the best score for each model.
| # | Model | Org | Elo |
|---|---|---|---|
| 1 | Gemini 4 Carbon | 1528 | |
| 2 | Gemini 4 Argon | 1520 | |
| 3 | Claude Opus 5.5 | Anthropic | 1509 |
| 4 | Muse Wonder | Meta | 1507 |
| 5 | Claude Fable 5.1 | Anthropic | 1504 |
| 6 | Claude Fable 5 | Anthropic | 1504 |
| 7 | Claude Opus 4.6 | Anthropic | 1503 |
| 8 | Claude Opus 4.7 | Anthropic | 1501 |
Elo is relative — a 20-point gap is small; only large gaps are meaningful.
Open-weight models
Highest open-weight models on the same blind-vote board, so the Elo is comparable with the frontier list.
| # | Model | Org | Elo |
|---|---|---|---|
| 1 | Kimi K3 | Moonshot | 1488 |
| 2 | GLM 5.3 | Z.ai | 1481 |
| 3 | MiMo V2.6 Pro | Xiaomi | 1480 |
| 4 | GLM 5.3 Flash | Z.ai | 1477 |
| 5 | DeepSeek V4.1 Flash 20260910 | DeepSeek | 1476 |
| 6 | GLM 5.2 | Z.ai | 1475 |
| 7 | MiMo V2.5 Pro | Xiaomi | 1467 |
| 8 | GLM 5.1 | Z.ai | 1465 |
Same Elo scale as the frontier board. Proprietary models are excluded; a 20-point gap is still small.
Source: LMArena · open weights ↗
Most used on OpenRouter
What developers actually run: models ranked by tokens processed through OpenRouter over the last day.
| # | Model | Org | T tokens |
|---|---|---|---|
| 1 | deepseek-v4.1-flash | deepseek | 39.57 |
| 2 | glm-5.3-flash | z-ai | 15.0 |
| 3 | gpt-6-luna | openai | 9.59 |
| 4 | hy4-preview | tencent | 8.96 |
| 5 | claude-opus-5.5 | anthropic | 7.86 |
| 6 | step-5-preview | stepfun | 5.39 |
| 7 | glm-5.3 | z-ai | 5.15 |
| 8 | nemotron-3-ultra-550b-a55b | nvidia | 4.44 |
| 9 | mimo-v2.6-flash | xiaomi | 4.1 |
| 10 | deepseek-v4-flash | deepseek | 4.07 |
Trillions of tokens processed. Usage, not quality — cheap and free models rank high here.
Open-weight video models
Open-weight video generators you can download and run, ranked by downloads across every variant of each model.
| # | Model | Org | M downloads |
|---|---|---|---|
| 1 | Wan2.2 | Wan-AI | 11.86 |
| 2 | LTX-2.5 | Lightricks | 4.43 |
| 3 | MiniMax-H3 | larryvrh | 2.5 |
| 4 | LTX-2.3 | Lightricks | 2.2 |
| 5 | Wan2.1 | Wan-AI | 1.76 |
| 6 | LTX-Video | Lightricks | 1.0 |
| 7 | Cobijada-Minimax-H3 | Winnougan | 0.4 |
| 8 | LTX-2 | Lightricks | 0.34 |
| 9 | Vdn-Minimax-H3 | t8star | 0.25 |
| 10 | FastVideo-FastH3 | FastVideo | 0.23 |
Millions of downloads, summing quantised and repackaged variants into their base model.