Model Stability Report
Which AI models are the most consistent over time? This report analyzes rank changes, state classifications, and sparkline volatility across 300 tracked models to produce a stability score from 0 to 100.
Rock Solid
32
Consistent
29
Variable
87
Volatile
152
Stability Classification Distribution
Provider Stability Rankings (Avg Score)
Most Stable Models
Top 20 models with the highest stability scores. These models maintain consistent rankings with minimal volatility.
| # | Model | Score | Stability | 24h | 7d |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.1Anthropic | 95.9 | 100 | 0 | 0 |
| 2 | Claude Fable 5.1 (batch)Anthropic | 95.9 | 100 | 0 | 0 |
| 3 | Claude Fable 5Anthropic | 95.9 | 100 | 0 | 0 |
| 4 | Claude Fable 5 (batch)Anthropic | 95.9 | 100 | 0 | 0 |
| 5 | Claude Opus 4.8Anthropic | 95.1 | 100 | -1 | 0 |
| 6 | Claude Opus 4.8 (batch)Anthropic | 94.6 | 99 | -1 | -2 |
| 7 | GPT-5.5 Pro (batch)OpenAI | 92.7 | 96 | -1 | -3 |
| 8 | GPT-5.5 (batch)OpenAI | 92.7 | 96 | -1 | -3 |
| 9 | Claude Opus 4.7 (batch)Anthropic | 95.1 | 96 | -1 | -3 |
| 10 | Gemini 3.1 Pro Preview (batch)Google | 92.2 | 96 | -1 | -3 |
| 11 | GPT-5.4 Pro (batch)OpenAI | 91.9 | 96 | -1 | -3 |
| 12 | GPT-5.4 (batch)OpenAI | 91.9 | 96 | -1 | -3 |
| 13 | Grok 4.6xAI | 88.8 | 96 | -1 | +3 |
| 14 | GPT-5.2 Pro (batch)OpenAI | 90.5 | 93 | -1 | -4 |
| 15 | GPT-5.2 (batch)OpenAI | 90.5 | 93 | -1 | -4 |
| 16 | GPT-5.6 Luna Pro (batch)OpenAI | 89.0 | 93 | -1 | -4 |
| 17 | GPT-5.6 Luna (batch)OpenAI | 89.0 | 93 | -1 | -4 |
| 18 | GPT-5.6 Terra Pro (batch)OpenAI | 89.0 | 93 | -1 | -4 |
| 19 | GPT-5.6 Terra (batch)OpenAI | 89.0 | 93 | -1 | -4 |
| 20 | GPT-5.6 Sol Pro (batch)OpenAI | 89.0 | 93 | -1 | -4 |
Most Volatile Models
Bottom 20 models with the lowest stability scores. These models show significant ranking fluctuations or inconsistent states.
| # | Model | Score | Stability | 24h | 7d |
|---|---|---|---|---|---|
| 1 | Gemini 3.6 Flash (batch)Google | 40.0 | 19 | -12 | -16 |
| 2 | Gemini 3.6 FlashGoogle | 40.0 | 19 | -12 | -16 |
| 3 | Command R+ (08-2024)Cohere | 48.7 | 34 | -2 | -9 |
| 4 | Command ACohere | 50.8 | 34 | -2 | -9 |
| 5 | Claude 3 HaikuAnthropic | 51.3 | 34 | -2 | -9 |
| 6 | Llama 3.1 8B InstructMeta | 44.5 | 37 | -2 | -9 |
| 7 | R1 Distill Llama 70BDeepSeek | 40.7 | 39 | -2 | -9 |
| 8 | Laguna S 2.1 (free)poolside | 40.0 | 39 | -12 | -16 |
| 9 | Laguna S 2.1poolside | 40.0 | 39 | -12 | -16 |
| 10 | Ling 3.0 Flashinclusionai | 40.0 | 39 | -12 | -16 |
| 11 | Claude Opus 5 (batch)Anthropic | 40.0 | 39 | -12 | -16 |
| 12 | Qwen3.7 FlashAlibaba | 40.0 | 39 | -12 | -17 |
| 13 | DeepSeek V4 Flash 0731DeepSeek | 40.0 | 39 | -13 | -18 |
| 14 | DeepSeek V4 Flash Latest~deepseek | 40.0 | 39 | -13 | -18 |
| 15 | Muse Glimmer 30Bmeta | 40.0 | 39 | -14 | -20 |
| 16 | Solar Pro 4Upstage | 40.0 | 39 | -14 | -20 |
| 17 | Sakana Namazusakana | 40.0 | 39 | -14 | -20 |
| 18 | Nemotron 3.5 Lightning (free)NVIDIA | 40.0 | 39 | -14 | -20 |
| 19 | Nemotron 3.5 LightningNVIDIA | 40.0 | 39 | -14 | -20 |
| 20 | LFM2.5-2.6B (free)Liquid AI | 40.0 | 39 | -14 | -20 |
Stability by Provider
Aggregated stability metrics per provider. Providers are ranked by their average stability score across all models.
| Provider | Models | Avg Stability |
|---|---|---|
| xAI | 7 | 69.0 |
| Anthropic | 27 | 68.4 |
| OpenAI | 87 | 65.5 |
| thinkingmachines | 4 | 62.6 |
| Inception | 2 | 59.9 |
| Amazon | 1 | 59.5 |
| IBM | 1 | 58.8 |
| Xiaomi | 5 | 57.3 |
| Zhipu AI | 18 | 56.5 |
| Moonshot AI | 2 | 55.4 |
| aion-labs | 2 | 54.0 |
| MiniMax | 7 | 53.3 |
| 28 | 52.1 | |
| Alibaba | 33 | 49.8 |
| arcee-ai | 1 | 48.5 |
| Mistral AI | 5 | 48.0 |
| StepFun | 1 | 47.7 |
| Cursor | 2 | 46.9 |
| Upstage | 2 | 46.5 |
| DeepSeek | 14 | 46.4 |
| Tencent | 6 | 46.4 |
| Microsoft | 1 | 44.0 |
| Meta | 5 | 42.8 |
| meta | 6 | 42.2 |
| prism-ml | 1 | 39.0 |
| unbiased | 1 | 39.0 |
| ~deepseek | 3 | 39.0 |
| inference-net | 2 | 39.0 |
| ~openai | 4 | 39.0 |
| sakana | 3 | 39.0 |
| inclusionai | 5 | 39.0 |
| ~z-ai | 2 | 39.0 |
| dots-studio | 1 | 39.0 |
| ByteDance | 2 | 39.0 |
| Liquid AI | 1 | 39.0 |
| NVIDIA | 2 | 39.0 |
| poolside | 2 | 39.0 |
| Cohere | 4 | 37.8 |
Stability Distribution
How stability scores are distributed across all 300 tracked models.
What Makes a Model Stable?
Our stability scoring system uses three key signals to measure how consistently a model performs over time.
Rank Consistency
The most direct measure of stability. Models lose up to 25 points for large 24-hour rank changes (5 points per rank position moved) and up to 21 points for 7-day changes (3 points per position). Models that hold their rank tightly score higher.
State Classification
Each model has a state reflecting its overall reliability. Models in a "stable" state receive a 10-point bonus, while "fragile" models are penalized 15 points. This captures systemic reliability beyond simple rank movement.
Sparkline Volatility
The 14-day sparkline data reveals hidden volatility. We compute the standard deviation of the sparkline and subtract up to 20 points. Even models that end where they started can be penalized if they oscillated wildly along the way.
Related
The stability score starts at 100 and is reduced based on three factors: 24-hour rank changes (up to -25 points, at 5 per position moved), 7-day rank changes (up to -21 points, at 3 per position), and sparkline volatility measured by standard deviation (up to -20 points). Models in a "stable" state get a +10 bonus, while "fragile" models lose 15 points.
Models are classified into four tiers based on their stability score: "Rock Solid" (85-100) means extremely consistent performance with minimal fluctuation. "Consistent" (70-84) means generally reliable with minor variations. "Variable" (50-69) shows noticeable ranking fluctuations. "Volatile" (below 50) indicates significant instability and unpredictable performance.
Stability indicates how predictably a model will perform over time. A highly rated but volatile model may deliver inconsistent results, which is problematic for production applications requiring reliable output quality. Stable models provide more predictable performance, making them safer choices for mission-critical workloads even if they do not always hold the top rank.