Model rankings
AI model rankings
See which models developers run through LLMFlux, measured from real gateway traffic, with tokens processed, request volume, success rate and latency. When a period has too little traffic to rank fairly, we show featured models from the live catalogue instead and label them that way.
Tokens processed36Mthe last 7 days
Requests721successful requests
Success rate84%across all requests
Top modelClaude Opus 5.547.2% of tokens
Token usage by model
Tokens processed per day over the last 7 days, stacked by model. Hover a bar for the split.
Nov 19 · busiest19M tokens
- Claude Opus 5.557%
- Other models43%
- Claude Opus 5.5
- DeepSeek Unrestricted
- DeepSeek V4.1 Flash
- Other models
Leaderboard
Ranked by tokens processed over the last 7 days. Change compares with the previous period of the same length.
| Model | Share | Details | |||
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 claude-opus-5-5 · Anthropic | 17M | +999%+ | Details | |
| 2 | DeepSeek Unrestricted deepseek-unrestricted · DeepSeek | 8.3M | New | Details | |
| 3 | DeepSeek V4.1 Flash deepseek-v4.1-flash · DeepSeek | 2.2M | New | Details | |
| · | Claude Fable 5 claude-fable-5 · Anthropic | Limited | — | — | Details |
| · | Claude Opus 5 claude-opus-5 · Anthropic | Limited | — | — | Details |
| · | DeepSeek V4-FLASH deepseek-v4-flash · DeepSeek | Limited | — | — | Details |
| · | GLM-5.3 Flash glm-5.3-flash · Zhipu AI | Limited | — | — | Details |
| · | Nemotron 3.5 Lightning nemotron-3.5-lightning · NVIDIA | Limited | — | — | Details |
Share by model maker
Share of tokens processed over the last 7 days.
- Anthropic69.8%
- DeepSeek29.4%
- Zhipu AI0.8%
Methodology
- Source. Every request through the LLMFlux gateway. Tokens and requests count successful requests (HTTP status below 400) only, so failed attempts are not counted as usage.
- Ranking. By tokens processed in the period. Change compares with the previous period of the same length.
- Quality. Success rate = successful ÷ all requests. Latency = median end-to-end time of successful requests.
- Privacy. Only aggregated totals are published. A model used by fewer than 2 customers shows “Limited” and gets no rank.
- Thresholds. Usage rankings need 1.0M tokens from 3 customers in the period; otherwise featured models are shown.
- Freshness. Last updated Nov 20, 2026, 3:46 PM UTC. Recalculated every 10 minutes.
Questions
- How are models ranked?
- When there is enough real traffic in the selected period, models are ranked by tokens processed through the LLMFlux gateway in successful requests (HTTP status below 400). Request counts, success rate and median latency are measured from the same requests.
- What does "Featured models" mean?
- When a period has too little traffic to rank fairly, the page lists models from the live catalogue in an order chosen by the LLMFlux team instead. That order is editorial, not a popularity ranking, and any usage figures shown next to it are still real.
- Why do some models show "Limited"?
- To protect customer privacy, usage figures are hidden for a model that only one or two customers used in the period, and it is not given a rank position.
- How often is the data updated?
- Rankings are recalculated on the server every few minutes. The time of the last update is shown on the page.
- Can I use every ranked model through one API?
- Yes. Every model listed is in the LLMFlux catalogue and can be called with one API key through the OpenAI-compatible or Anthropic-compatible endpoints.
Try any ranked model with one API key
OpenAI- and Anthropic-compatible endpoints, streaming, and per-request usage logs.