#1
General use
DeepSeek V4 Flash
12.1M tokens
Rolling 8-week window
AI model rankings
Real, aggregate usage signals from the NexoRouter gateway. The ranking does not claim absolute quality; it shows which models are used inside a defined window.
Top models
Tokens from ranked models by week.
DeepSeek V4 FlashGLM 4.6Claude Opus 4.8GLM 5.1O4 MiniQwen3-VL-PlusGPT-4o-Mini
Leaderboard
| Rank | Model | Requests | Tokens | Share | Trend |
|---|---|---|---|---|---|
| 01 | DeepSeek V4 Flash DeepSeek | 801 | 12.1M | 82.3% | +78122% |
| 02 | GLM 4.6 ZAI | 45 | 1.6M | 10.7% | — |
| 03 | Claude Opus 4.8 Anthropic | 44 | 916.9K | 6.2% | — |
| 04 | GLM 5.1 Zhipu AI | 2 | 38.9K | 0.3% | — |
| 05 | O4 Mini OpenAI | 2 | 33.6K | 0.2% | — |
| 06 | Qwen3-VL-Plus Alibaba Cloud | 33 | 24.1K | 0.2% | — |
| 07 | GPT-4o-Mini OpenAI | 611 | 8.6K | 0.1% | +31% |
| 08 | Kimi K3 Moonshot AI | 35 | 4.5K | 0.0% | — |
| 09 | Qwen3-Max Alibaba Cloud | 4 | 2.3K | 0.0% | -52% |
| 10 | Kimi K2.6 Moonshot AI | 28 | 2K | 0.0% | — |
Top models by task
Directional grouping built from the model’s published type and recent usage; this is not a quality benchmark.
#16
Coding
Kimi K2.7 Code
227 tokens
Reasoning
Insufficient sample
#6
Multimodal
Qwen3-VL-Plus
24.1K tokens
Reliability signal
Aggregate completed-request rate compared with error records in the same window.
Methodology: only aggregate gateway records are used. Prompts, identities, keys, and individual logs are never published. Percentages may move with small samples and should not be read as a performance guarantee.