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
45
Consistent
70
Variable
56
Volatile
129
Top 20 models with the highest stability scores. These models maintain consistent rankings with minimal volatility.
| # | Model | Score | Stability | 24h | 7d |
|---|---|---|---|---|---|
| 1 | GPT-5.4OpenAI | 94.0 | 100 | 0 | -1 |
| 2 | Llemma 7beleutherai | 47.5 | 100 | 0 | -1 |
| 3 | GPT-3.5 Turbo (older v0613)OpenAI | 38.0 | 100 | -1 | 0 |
| 4 | Inflection 3 ProductivityInflection | 36.8 | 100 | +1 | 0 |
| 5 | MellumJetBrains | 32.6 | 100 | +1 | 0 |
| 6 | QwQ 32BAlibaba | 47.1 | 100 | 0 | -3 |
| 7 | Seed 1.6 FlashByteDance | 85.0 | 100 | 0 | -3 |
| 8 | Olmo 3 7B InstructAllen AI | 69.0 | 99 | 0 | -3 |
| 9 | Mistral Large 2411Mistral AI | 49.9 | 99 | 0 | -3 |
| 10 | Gemma 3n 4BGoogle | 46.3 | 98 | 0 | +3 |
| 11 | Command R+ (08-2024)Cohere | 47.8 | 98 | 0 | +3 |
| 12 | Llama 3.2 3B Instruct (free)Meta | 35.3 | 98 | -2 | 0 |
| 13 | GPT-5.4 ProOpenAI | 94.0 | 97 | 0 | +3 |
| 14 | Granite 4.0 MicroIBM | 55.1 | 96 | -2 | -1 |
| 15 | GPT-3.5 Turbo 16kOpenAI | 39.9 | 96 | -1 | +2 |
| 16 | Nemotron Nano 12B 2 VLNVIDIA | 72.6 | 96 | +1 | +2 |
| 17 | Nemotron 3 Super (free)NVIDIA | 84.1 | 96 | +1 | +2 |
| 18 | Llama 3.1 405B (base)Meta | 38.7 | 96 | -2 | -1 |
| 19 | GPT-4o (2024-05-13)OpenAI | 52.7 | 96 | -2 | -1 |
| 20 | Qwen2.5 VL 32B InstructAlibaba | 56.8 | 95 | +1 | -3 |
Bottom 20 models with the lowest stability scores. These models show significant ranking fluctuations or inconsistent states.
| # | Model | Score | Stability | 24h | 7d |
|---|---|---|---|---|---|
| 1 | Devstral 2 2512Mistral AI | 67.7 | 35 | -12 | +10 |
| 2 | MiniMax M1MiniMax | 68.5 | 35 | -10 | +15 |
| 3 | Claude Sonnet 4Anthropic | 79.9 | 35 | -10 | +19 |
| 4 | UI-TARS 7B ByteDance | 62.7 | 35 | +15 | +8 |
| 5 | GPT-4 TurboOpenAI | 60.5 | 35 | -10 | +10 |
| 6 | Molmo2 8BAllen AI | 67.6 | 35 | -9 | +15 |
| 7 | Mistral Large 3 2512Mistral AI | 73.5 | 35 | +19 | +9 |
| 8 | MiniMax M2.5 (free)MiniMax | 83.4 | 35 | +11 | +11 |
| 9 | LFM2.5-1.2B-Thinking (free)Liquid AI | 59.0 | 35 | +7 | +12 |
| 10 | LFM2-24B-A2BLiquid AI | 53.2 | 35 | -7 | +9 |
| 11 | gpt-oss-20b (free)OpenAI | 73.9 | 35 | +21 | +17 |
| 12 | o4 Mini Deep ResearchOpenAI | 85.0 | 36 | -5 | +16 |
| 13 | GPT-5 NanoOpenAI | 75.7 | 36 | -5 | +22 |
| 14 | Seed 1.6ByteDance | 85.0 | 36 | -7 | +10 |
| 15 | Seed-2.0-MiniByteDance | 85.0 | 36 | -7 | +7 |
| 16 | Seed-2.0-LiteByteDance | 85.0 | 36 | -8 | +11 |
| 17 | GPT-4.1 NanoOpenAI | 80.8 | 36 | +12 | +14 |
| 18 | GPT-5.3 ChatOpenAI | 85.0 | 36 | +30 | +16 |
| 19 | Qwen Plus 0728Alibaba | 77.0 | 36 | -8 | +18 |
| 20 | GPT Audio MiniOpenAI | 68.4 | 36 | +10 | +10 |
Aggregated stability metrics per provider. Providers are ranked by their average stability score across all models.
| Provider | Models | Avg Stability |
|---|---|---|
| eleutherai | 1 | 100.0 |
| JetBrains | 1 | 100.0 |
| Inflection | 2 | 96.5 |
| IBM | 1 | 96.1 |
| Windsurf | 1 | 82.8 |
| essentialai | 1 | 77.8 |
| AI21 Labs | 1 | 75.7 |
| Microsoft | 1 | 74.4 |
| Cohere | 4 | 74.3 |
| Allen AI | 7 | 71.2 |
| aion-labs | 3 | 69.5 |
| Vercel | 1 | 69.5 |
| Meta | 13 | 65.4 |
| Moonshot AI | 4 | 65.1 |
| Amazon | 5 | 64.8 |
| 23 | 63.0 | |
| OpenAI | 59 | 62.7 |
| Anthropic | 13 | 60.0 |
| MiniMax | 8 | 60.0 |
| Perplexity | 5 | 59.6 |
| Mistral AI | 25 | 59.4 |
| NVIDIA | 11 | 59.4 |
| Baidu | 5 | 56.9 |
| Alibaba | 51 | 55.8 |
| arcee-ai | 7 | 54.0 |
| xAI | 10 | 50.8 |
| Meituan | 1 | 50.6 |
| Liquid AI | 5 | 49.4 |
| DeepSeek | 11 | 49.4 |
| ByteDance | 5 | 48.3 |
| Xiaomi | 3 | 46.2 |
| Kuaishou | 1 | 45.1 |
| Cursor | 2 | 45.1 |
| Inception | 3 | 38.2 |
| Upstage | 1 | 38.2 |
| Tencent | 1 | 37.5 |
| StepFun | 2 | 37.4 |
| deepcogito | 1 | 36.0 |
| Writer | 1 | 35.9 |
How stability scores are distributed across all 300 tracked models.
Our stability scoring system uses three key signals to measure how consistently a model performs over time.
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.
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.
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.
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.