DeepSeek
The most-used model lab across the full 63 weeks, finishing 15.8% ahead of StepFun.
20.63T
tokens across 63 weeks
Back to Leaderboard
AI Provider Token Race
Watch how AI provider usage evolved week by week, from Jun 2025 – Sep 2026. See the rise and fall of different providers as developers shift their preferences.
Week of Jun 30, 2025
14.0B tokens across all labs
01Google
8.8B
02Anthropic
4.7B
03DeepSeek
201.0M
04OpenAI
121.8M
05Mistral
64.7M
06Alibaba
28.5M
07MMeta
15.8M
08BBaidu
9.5M
09MoonshotAI
7.5M
10NNousResearch
3.9M
Token usage across Kilo Code
DeepSeekStepFunMiniMaxAnthropicZ.aiInclusionAIOpenAIPoolsidexAIAlibabaGoogleNVIDIAXiaomiMoonshotAITencentMeituanMistralNNexAGIByteDanceKwaipilotAArceeMMetaMMorphPPerplexityAAmazonNNousResearchBBaidu
Open-weight adoption
How open models gained ground
Follow open-weight share over time, then see which labs supplied the usage behind each week.
Latest share
65.3%
Change since Jun 30, 2025
+63.0 pts
Latest open-weight volume
6.27T
Verified releasesQwen3 Coder 480B-A35B InstructJul 22GLM 5.2Jun 17Laguna S 2.1Jul 21Kimi K3Jul 27
Share of all token usage
Open-weight versus proprietary models
Open weight Proprietary Total volume
Open-weight usage by lab
Top 10 labs across the period; all remaining usage is retained as other
What 63 weeks of model usage tell us
Signals calculated from 63 complete weeks of real Kilo developer token usage.
Usage Champion
View lab
xAI
Held the number one position for 17 consecutive weeks, the longest run of any lab.
17 consecutive weeks
17 total weeks at #1
Anthropic
No lab spent more weeks among the top three, while remaining in the top ten for 63 of 63 weeks.
39 weeks
ranked in the top 3
StepFun
Weekly usage rose from 3.6B to 1.27T after first crossing 1B tokens on Feb 2, 2026.
348x growth
1.26T weekly tokens gained
Top open-weight model labs by usage
Open-weight usage ranked by tokens across the full period, based on each model's licensing rather than its lab.
- #1
DeepSeek
1.70T tokens in the latest complete week.
View models20.63T63-week usage - #2
StepFun
1.27T tokens in the latest complete week.
View models17.82T63-week usage - #3
MiniMax
281.1B tokens in the latest complete week.
View models15.05T63-week usage
Open-weight releases that drove adoption
57 releases showed >20% week-over-week growth. These are the ones that moved the needle most.
- #1
DeepSeek: DeepSeek V4 Flash 0731
Released Jul 31. Usage jumped from 938.2B to 1.81T tokens the following week.
View lab+92%+867.0B tokens - #2
Tencent: Hy3
Released Jul 6. Usage jumped from 692.2M to 718.4B tokens the following week.
View lab+103683%+717.7B tokens - #3
inclusionAI: Ling 3.0 Flash
Released Jul 23. Usage jumped from 800.4B to 1.24T tokens the following week.
View lab+55%+441.7B tokens - #4
Z.ai: GLM 5
Released Feb 11. Usage jumped from 237.2B to 590.0B tokens the following week.
View lab+149%+352.8B tokens
Real usage from real developers
This animation shows weekly token consumption by AI provider across all Kilo Code users. Each bar represents the total tokens processed by models from that provider during that week.
- Time Range
- Jun 2025 – Sep 2026
- Granularity
- Weekly snapshots
- Metric
- Total tokens consumed
How to read the race
Watch the rankings
Providers move up and down as their weekly usage changes relative to others; use the filters to compare how open-weight and proprietary model usage evolves over time.
Compare bar lengths
Bar length shows absolute token usage — longer bars mean more tokens consumed.
Use the controls
Play, pause, scrub through time, or adjust speed to explore at your own pace.
Share the signals
The insight cards summarize leadership, breakout growth, open-weight labs, and market concentration.