模型新鲜度追踪器
分析AI模型及其数据的最近更新时间。追踪新鲜度评分，识别过时数据，查看哪些服务商的模型保持最新。
100.0
满分100
300
个模型已更新
300
个模型已更新
Claude Fable 5
0m ago
模型新鲜度分布
新鲜度层级分布
按数据最后更新时间对模型进行分组。
最新鲜的模型
Top 20数据最近更新的模型，按新鲜度评分排名。
| # | 模型 | 提供商 | 最后更新 | 新鲜度 |
|---|---|---|---|---|
| 1 | Falcon-H1-Arabic 3B Instruct | TII | 0m ago | 100 |
| 2 | Seed 1.6 Flash | ByteDance | 0m ago | 100 |
| 3 | Seed 1.6 | ByteDance | 0m ago | 100 |
| 4 | Nemotron 3 Nano 30B A3B | NVIDIA | 0m ago | 100 |
| 5 | Nemotron 3 Nano 30B A3B (free) | NVIDIA | 0m ago | 100 |
| 6 | Nemotron 3 Super | NVIDIA | 0m ago | 100 |
| 7 | Nemotron 3 Super (free) | NVIDIA | 0m ago | 100 |
| 8 | Seed-2.0-Lite | ByteDance | 0m ago | 100 |
| 9 | Seed-2.0-Mini | ByteDance | 0m ago | 100 |
| 10 | Aion-2.0 | aion-labs | 0m ago | 100 |
| 11 | Qwen3 Coder Next | Alibaba | 0m ago | 100 |
| 12 | Solar Pro 3 | Upstage | 0m ago | 100 |
| 13 | Palmyra X5 | Writer | 0m ago | 100 |
| 14 | GPT Audio | OpenAI | 0m ago | 100 |
| 15 | GPT Audio Mini | OpenAI | 0m ago | 100 |
| 16 | Falcon-H1-Arabic 34B Instruct | TII | 0m ago | 100 |
| 17 | Falcon-H1-Arabic 7B Instruct | TII | 0m ago | 100 |
| 18 | MoonshotAI Kimi Latest | ~moonshotai | 0m ago | 100 |
| 19 | Google Gemini Flash Latest | 0m ago | 100 | |
| 20 | Anthropic Claude Sonnet Latest | ~anthropic | 0m ago | 100 |
最陈旧的模型
Bottom 20数据最陈旧的模型。这些模型可能有过时的评分或缺失的信号更新。
| # | 模型 | 提供商 | 最后更新 | 新鲜度 |
|---|---|---|---|---|
| 1 | Claude Fable 5 | Anthropic | 0m ago | 100 |
| 2 | Claude Fable 5 (batch) | Anthropic | 0m ago | 100 |
| 3 | Claude Opus 5 (Fast) | Anthropic | 0m ago | 100 |
| 4 | Claude Opus 5 | Anthropic | 0m ago | 100 |
| 5 | Claude Opus 4.8 (Fast) | Anthropic | 0m ago | 100 |
| 6 | Claude Opus 4.8 | Anthropic | 0m ago | 100 |
| 7 | Claude Opus 4.7 (Fast) | Anthropic | 0m ago | 100 |
| 8 | Claude Opus 4.7 | Anthropic | 0m ago | 100 |
| 9 | Claude Opus 4.7 (batch) | Anthropic | 0m ago | 100 |
| 10 | Claude Opus 4.8 (batch) | Anthropic | 0m ago | 100 |
| 11 | GPT-5.5 Pro | OpenAI | 0m ago | 100 |
| 12 | GPT-5.5 Pro (batch) | OpenAI | 0m ago | 100 |
| 13 | GPT-5.5 | OpenAI | 0m ago | 100 |
| 14 | GPT-5.5 (batch) | OpenAI | 0m ago | 100 |
| 15 | Gemini 3.1 Pro Preview Custom Tools | 0m ago | 100 | |
| 16 | Gemini 3.1 Pro Preview | 0m ago | 100 | |
| 17 | Gemini 3.1 Pro Preview (batch) | 0m ago | 100 | |
| 18 | GPT-5.4 Pro | OpenAI | 0m ago | 100 |
| 19 | GPT-5.4 Pro (batch) | OpenAI | 0m ago | 100 |
| 20 | GPT-5.4 | OpenAI | 0m ago | 100 |
各服务商更新频率
每个服务商的数据有多新鲜？按所有模型的平均新鲜度评分排序。
| 提供商 | 模型 | 平均评分 |
|---|---|---|
| ByteDance | 4 | 100.0 |
| TII | 3 | 100.0 |
| Upstage | 1 | 100.0 |
| Writer | 1 | 100.0 |
| NVIDIA | 9 | 100.0 |
| ~moonshotai | 1 | 100.0 |
| rekaai | 1 | 100.0 |
| ~anthropic | 4 | 100.0 |
| ~openai | 2 | 100.0 |
| 2 | 100.0 | |
| aion-labs | 3 | 100.0 |
| perceptron | 1 | 100.0 |
| Kuaishou | 3 | 100.0 |
| inclusionai | 5 | 100.0 |
| poolside | 4 | 100.0 |
| Meituan | 1 | 100.0 |
| ~x-ai | 1 | 100.0 |
| sakana | 1 | 100.0 |
| StepFun | 2 | 100.0 |
| Cohere | 4 | 100.0 |
| Moonshot AI | 8 | 100.0 |
| Microsoft | 1 | 100.0 |
| IBM | 1 | 100.0 |
| Allen AI | 1 | 100.0 |
| ~deepseek | 1 | 100.0 |
| Mistral AI | 6 | 100.0 |
| Meta | 5 | 100.0 |
| Alibaba | 31 | 100.0 |
| Cursor | 2 | 100.0 |
| arcee-ai | 1 | 100.0 |
| Inception | 1 | 100.0 |
| Amazon | 1 | 100.0 |
| DeepSeek | 12 | 100.0 |
| thinkingmachines | 3 | 100.0 |
| Zhipu AI | 13 | 100.0 |
| MiniMax | 8 | 100.0 |
| Tencent | 2 | 100.0 |
| Xiaomi | 2 | 100.0 |
| 27 | 100.0 | |
| OpenAI | 86 | 100.0 |
| xAI | 5 | 100.0 |
| meta | 2 | 100.0 |
| Anthropic | 28 | 100.0 |
方法论
新鲜度评分
计算方式为100减去自上次更新以来的小时数，上限为100小时。100分表示模型刚刚更新；0分表示超过100小时（4天以上）未更新。
新鲜度层级
模型根据更新时间分为五个层级：刚刚更新（<1小时）、新鲜（1-6小时）、近期（6-24小时）、老化（1-7天）和过时（>7天）。没有已知更新时间戳的模型被归类为过时。
为什么新鲜度很重要
过时的数据可能导致不准确的排名和过时的信号评分。数据新鲜的模型反映最新的价格、功能和基准测试结果。监控新鲜度有助于及早发现数据管道问题。
Data freshness measures how recently each AI model's information has been updated. A freshness score of 100 means the model was just updated, while 0 means its data is over 100 hours (4+ days) old.
Our data pipeline refreshes model information hourly from multiple API sources. Each update pulls the latest pricing, capabilities, and benchmark data for all tracked models.
Stale data can lead to inaccurate rankings and outdated signal scores. Models with fresh data reflect the latest pricing changes, capability updates, and benchmark results, ensuring you make decisions based on current information.
Models are grouped into five tiers: Just Updated (under 1 hour), Fresh (1-6 hours), Recent (6-24 hours), Aging (1-7 days), and Stale (over 7 days). Models without a known update timestamp are classified as Stale.