CodeSOTAIntelligence

OpenRouter Usage & Churn

ort.fabryka.ai · updated 2026-09-25 02:14 UTC · market memo → · vendor migration → · normalized → · apps / demand → · east vs west →
686 days of history 2024-11-09 → 2026-09-25 865 models tracked

Top vendors — daily tokens, 7-day avg (full history, billions)

Substitution — Claude Opus version migration (stacked, 7-day avg, billions)

Each band is one version. Watch usage hand off up the stack: 4.5→4.6→4.7→4.8 — the family total holds/grows while users rotate to the newest release (in-family retention, not churn).

Vendor churn & market share (30d vs 30d, tokens)

vendorrecentΔ%shareshare Δ
deepseek110.9T+44%24.0%-4.3pp
z-ai77.6T+353%16.8%+10.5pp
openai74.4T+129%16.1%+4.1pp
tencent73.0T+120%15.8%+3.6pp
xiaomi29.2T-5%6.3%-5.0pp
google28.8T+13%6.2%-3.1pp
anthropic21.1T-4%4.6%-3.5pp
moonshotai8.1T+7%1.8%-1.0pp
meta7.8Tnew1.7%+1.5pp
qwen7.0T+63%1.5%-0.1pp
minimax6.7T-19%1.5%-1.6pp
upstage6.1T+489%1.3%+0.9pp
x-ai2.5T-0%0.5%-0.4pp
mistralai1.2T+12%0.3%-0.1pp

Family churn & substitution (30d vs 30d) — version migration nets out

familyrecentΔ%lead version (stickiness)substitution
DeepSeek DeepSeek Flash66.0T+8%DeepSeek V4 Flash 0731 71%vV4 0423 → vV4 0731
OpenAI GPT Luna (batch)53.5T+198%GPT-5.6 Luna (batch) 98%
Xiaomi MiMo25.6T-10%MiMo-V2.5 100%
Tencent Hy319.0T-43%Hy3 100%
nvidia/nemotron ultra 550b a5517.9T+34%Nemotron 3 Ultra 550B 100%
Google Gemini Flash (batch)16.1T+30%Gemini 3.8 Flash (batc 45%v3.6 → v3.8
Z.ai GLM (batch)11.5TnewGLM 5.3 (batch) 100%
DeepSeek DeepSeek Pro9.9T-24%DeepSeek V4 Pro 0423 53%vV4 0423 → vV4 0813
Z.ai GLM9.4T-41%GLM 5.2 (free) 93%v5.2 → v4.6
Anthropic Claude Sonnet (batch9.4T+12%Claude Sonnet 5 (batch 64%v4.6 → v5
Anthropic Claude Opus (batch)8.7T-23%Claude Opus 5 (batch) 66%v4.7 → v5.5

Model lifecycle distribution

stage#volumeshare of trackedexamples
LAUNCHING2510.6T/d
Ox Alpha, DeepSeek V4.1 Fl, Space Bunny Alph
GROWING71.0T/d
Owl Alpha, GLM 5.3 (batch), Qwen3.7 Flash
MATURE154.2T/d
GLM 5.3 Flash (b, Muse Spark 1.3 C, Solar Pro 4
DECLINING1296.9T/d
GPT-5.6 Luna (ba, DeepSeek V4 Flas, MiMo-V2.5
DEAD3997.9B/d
Gemini 3.6 Flash, Claude Opus 4.7 , MiniMax M2.7

📈 Gaining by spend (USD, 7d vs 7d)

modeltiernowΔ ($)
Claude Opus 4.8 (batch)paid$3.3M+$2.2M
MiniMax M3paid$892.9K+$876.2K
DeepSeek V4 Pro 0423paid$1.6M+$493.4K
Claude Opus 4.8 (Fast)paid$379.8K+$197.1K
Step 3.7 Flashpaid$189.6K+$149.5K
MiMo-V2.5paid$341.4K+$100.7K
Hy3 previewfree$616.0K+$61.3K
DeepSeek V4 Flash 0423paid$204.8K+$57.7K
GPT-5.1 (batch)paid$86.6K+$40.2K
Gemini 3 Flash Preview (batch)free$325.4K+$33.3K
Qwen3.5 397B A17Bpaid$61.8K+$26.5K
GPT-4o (2024-05-13)paid$27.1K+$22.4K

📉 Churning by spend (USD, 7d vs 7d)

modeltiernowΔ ($)
Claude Opus 4.7 (batch)paid$3.9M−$1.9M
MiMo-V2.5-Propaid$225.4K−$387.4K
Claude Opus 4.6 (batch)paid$1.2M−$265.9K
Claude Sonnet 4.6 (batch)paid$2.9M−$181.2K
Kimi K2.6paid$370.0K−$177.9K
Claude Opus 4.7 (Fast)paid$127.9K−$96.5K
MiniMax M2.7paid$98.3K−$74.6K
GLM 4.7paid$57.8K−$72.5K
Claude Opus 4.5 (batch)paid$120.7K−$51.3K
Claude Sonnet 4paid$135.7K−$38.8K
Qwen3.6 Pluspaid$37.8K−$34.8K
GLM 5.1paid$297.6K−$31.5K

📈 Long-term gainers (tokens, 30d vs 30d)

modeltiernowΔ (tokens)
GPT-5.6 Luna (batch)paid52.5T+34.5T
GLM 5.3 (batch)paid11.5T+10.7T
DeepSeek V4 Flash 0731paid47.0T+10.6T
Minimax M3free8.1T+8.0T
Solar Pro 4paid6.1T+5.0T
Nemotron 3 Ultra 550B A55Bfree17.9T+4.5T
GPT-5.6 Sol (batch)paid7.8T+4.2T
Gemini 3.7 Flash (batch)paid6.5T+3.1T
DeepSeek V4 Pro 0813paid4.7T+2.7T
Dots 3 Note Previewfree2.0T+1.8T
Claude Sonnet 5 (batch)paid6.0T+1.4T
Nemotron 3.5 Lightningfree3.4T+1.4T

📉 Long-term churn (tokens, 30d vs 30d)

modeltiernowΔ (tokens)
Ox Alphapaid4.0T−19.3T
Hy3paid19.0T−14.2T
GLM 5.2 (free)free8.7T−6.4T
Gemini 3.6 Flash (batch)paid288.7B−6.1T
DeepSeek V4 Flash 0423paid19.0T−6.0T
DeepSeek V4 Pro 0423paid5.2T−5.9T
Step 3.7 Flashpaid899.4B−3.7T
MiMo-V2.5paid25.6T−2.8T
Claude Opus 4.7 (batch)paid189.4B−1.2T
Claude Opus 5 (batch)paid5.8T−1.2T
Laguna S 2.1free5.3T−1.2T
MiMo-V2.5-Propaid1.2T−1.1T