Annual report · January to December 2025

State of LLM Usage 2025

Which models, labs and providers carried production traffic this year. Measured from real token volume routed through LLM Gateway, not from benchmarks or surveys.

See live rankings
Tokens processed
32.7B
Requests
5.8M
Models with traffic
83
Providers serving them
18

Key findings

The short version of 2025.

  • Gemini 2.5 Flash Preview (09-2025) was the most used model, carrying 92.3% of all tokens.
  • Google models processed 95.4% of token volume, more than any other lab.
  • December was the busiest month with 18.5B tokens.
  • 2 different models held the monthly number-one spot.
  • GLM-4.6 grew fastest, up 605,294% over the previous period.

Token volume by month

Total tokens processed each month, with the model that led it.

  1. September96.3MLed by GPT-5 Mini
  2. October167.4MLed by Gemini 2.5 Flash Preview (09-2025)
  3. November13.9BLed by Gemini 2.5 Flash Preview (09-2025)
  4. December18.5BLed by Gemini 2.5 Flash Preview (09-2025)

Most used LLMs of 2025

Ranked by tokens across every provider that serves the model.

  1. 1Gemini 2.5 Flash Preview (09-2025)30.1B · 92.3%
  2. 2GLM-4.61B · 3.2%
  3. 3Gemini 3 Pro Image (Preview)1B · 3.1%
  4. 4GPT-5 Mini140.8M · 0.4%
  5. 5GLM-4.5V93.6M · 0.3%
  6. 6DeepSeek V3.163.3M · 0.2%
  7. 7Claude Sonnet 4 (2025-05-14)32.7M · 0.1%
  8. 8GPT OSS 120B27.6M · 0.1%
  9. 9Claude Opus 4.521M · 0.1%
  10. 10Claude Sonnet 4.518.3M · 0.1%
  11. 11Gemini 3 Pro (Preview)17.8M · 0.1%
  12. 12GPT-5 Nano11.3M · 0.0%
  13. 13GPT-5.1 Codex mini11.3M · 0.0%
  14. 14GLM-4.5 Flash7.1M · 0.0%
  15. 15Grok 4 Fast Non-Reasoning5.4M · 0.0%
  16. 16GPT-4o5.3M · 0.0%
  17. 17Gemini 2.5 Pro4.4M · 0.0%
  18. 18GLM-4.6V Flash4.2M · 0.0%
  19. 19Qwen3 Coder 480B A35B Instruct2.9M · 0.0%
  20. 20Claude Haiku 4.52.6M · 0.0%
  21. 21GPT-5.22.4M · 0.0%
  22. 22MiniMax M21.8M · 0.0%
  23. 23GPT-51.5M · 0.0%
  24. 24Kimi K2 Thinking1.1M · 0.0%
  25. 25Claude 3.7 Sonnet (2025-02-19)1M · 0.0%

Share by model lab

Token volume grouped by the lab that built the model.

  1. 1Google31.2B · 95.4%
  2. 2Z AI1.1B · 3.5%
  3. 3OpenAI202.2M · 0.6%
  4. 4Anthropic77M · 0.2%
  5. 5DeepSeek63.5M · 0.2%
  6. 6xAI6.7M · 0.0%
  7. 7Alibaba Cloud3.5M · 0.0%
  8. 8MiniMax1.8M · 0.0%
  9. 9Moonshot AI1.2M · 0.0%
  10. 10Perplexity38.1k · 0.0%

Provider market share

Who actually served the tokens, including third-party hosts.

  1. 1Google Vertex AI30.1B · 92.4%
  2. 2Google AI Studio1.1B · 3.5%
  3. 3Z AI960.3M · 3.0%
  4. 4CanopyWave220.3M · 0.7%
  5. 5OpenAI67.8M · 0.2%
  6. 6Anthropic30.6M · 0.1%
  7. 7DeepSeek25.9M · 0.1%
  8. 8Groq13.2M · 0.0%
  9. 9AWS Bedrock7.2M · 0.0%
  10. 10xAI6.7M · 0.0%

Fastest-growing models

Growth in token volume over the most recent months of the period, against the months before them.

  1. +605,294%

    GLM-4.6

    170.8k to 1B tokens

  2. +30,494%

    Gemini 2.5 Flash Preview (09-2025)

    98.2M to 30B tokens

Methodology

How to read the numbers.

The report counts tokens and requests routed through LLM Gateway in 2025, in UTC, for models in the public catalogue. A model served by several providers is counted once in the model ranking and split by host in the provider ranking.

Usage shows where production workloads run. It is not a quality score: a model can top a benchmark and still see little real traffic. Gateway traffic also leans towards developers who route across providers, so read shares as a view of that audience rather than of the whole market.

Run every model in this report through one API

One OpenAI-compatible endpoint, one key, automatic failover between providers.

Start for freeBrowse models

Other years: 2026

FAQ

About this report

Can't find an answer? Contact us

Every figure is aggregated from requests routed through LLM Gateway during the calendar year, in UTC. Nothing is surveyed or estimated.