One year of OpenRouter: who actually won 2025→2026
OpenRouter's rankings page ships a full year of weekly token data per vendor. We analyzed it. The story is brutal: while benchmark leaderboards talk about Claude, GPT-5, and Gemini, the actual inference market has shifted somewhere else entirely. Chinese open-weight labs went from a rounding error to more than half of the tracked flow in twelve months, and the total market is up 11×.
11.1×
Market growth (12mo)
39×
Chinese labs token growth
52%
Chinese share now (from 15%)
53
Weeks tracked
Vendor market share, week-by-week
Stacked share of total OpenRouter top-9 token volume. This is the "how many tokens" view — it undercounts premium vendors whose tokens cost 10–30× more. See the spend-share chart below for the dollar shape.
What changed in 12 months
Chinese labs: 0 → dominant
Twelve months ago, Chinese labs made up 15% of tracked flow — almost all of it DeepSeek and a little Qwen. Today they're 52%. In absolute tokens that's 1.02T → 39.9T — roughly 39× growth. Xiaomi alone went from nonexistent to 13% share in under a year.
Western incumbents: absolute growth, relative collapse
Google and Anthropic didn't lose volume — both grew several-fold. But the market grew faster around them. Google's share fell from ~37% → ~13%. Anthropic from ~25% → ~12%. Meta, Mistral, and Microsoft vanished from the top-9 entirely. OpenAI held roughly flat at ~10% share. "Losing the market" and "making more money" are the same thing here.
Chinese labs share, weekly
Combined share of xiaomi, qwen, minimax, deepseek, z-ai, stepfun.
Biggest share shifts
First 4 weeks of the series vs the last 4 weeks. Growth ratio is absolute tokens, not share.
| Vendor | Year-ago share | Now share | Δ points | Year-ago tokens | Now tokens | Absolute growth |
|---|---|---|---|---|---|---|
| 37.0% | 13.3% | -23.7 | 2.56T | 10.15T | 4.0× | |
| xiaomiCN | 0.0% | 13.0% | +13.0 | 0 | 9.90T | new |
| qwenCN | 2.2% | 12.7% | +10.5 | 154.1B | 9.71T | 62× |
| anthropic | 24.7% | 12.3% | -12.3 | 1.71T | 9.44T | 5.5× |
| openai | 11.4% | 9.8% | -1.6 | 788.1B | 7.47T | 9.5× |
| minimaxCN | 0.0% | 9.5% | +9.5 | 0 | 7.24T | new |
| deepseekCN | 12.5% | 6.3% | -6.2 | 864.8B | 4.84T | 5.6× |
| z-aiCN | 0.0% | 6.0% | +6.0 | 0 | 4.58T | new |
| stepfunCN | 0.0% | 4.8% | +4.8 | 0 | 3.66T | new |
| x-ai | 0.4% | 1.0% | +0.6 | 25.2B | 735.7B | 29× |
| nvidia | 0.0% | 0.3% | +0.3 | 0 | 231.2B | new |
| meta-llama | 5.6% | 0.0% | -5.6 | 385.3B | 0 | -100% |
| mistralai | 2.9% | 0.0% | -2.9 | 197.7B | 0 | -100% |
| microsoft | 0.8% | 0.0% | -0.8 | 53.0B | 0 | -100% |
| nousresearch | 0.3% | 0.0% | -0.3 | 17.8B | 0 | -100% |
Is the market actually following the benchmark leaders?
The short answer: no, and the longer answer has three parts.
1. The market is stratifying, not choosing. Claude Opus 4.6 is the #1 model on our inverted leaderboard by dollar spend — $25M/month across 24 apps — because it's priced at $5/$25 per million. But it's only ~4th by token volume (2.4T), and Anthropic as a vendor is ~12% of total tokens, down from ~25% last year. Both things are true at once: the premium lane still pays Anthropic real money, and the commodity lane has moved decisively elsewhere. Benchmark leaders don't lose revenue — they lose the long tail of workloads they used to own.
2. It's following price, not quality. The vendors gaining share (Qwen, Xiaomi, MiniMax, DeepSeek, Z.ai, StepFun) share one property: aggressive open-weight pricing, often sub-$1/M blended. On the inverted model leaderboard, Qwen3.6 Plus is in 27 of 30 apps and MiMo-V2-Pro handles 5.5T tokens at ~$1.50/M blended — a tiny fraction of what a comparable Claude Opus run would cost.
3. Benchmark-to-spend correlation is weak and inverted at the top. Among models with both token volume and a published benchmark score, the relationship between benchmark rank and market share is closer to anti-correlated. Premium models capture most of the dollar spend because they're priced 10–30× higher, but they don't capture tokens. Agents — when given a free choice through a router — route most tokens to "good enough and 20× cheaper" rather than "best and 20× more expensive".
This cuts two ways. It says benchmark leaderboards overstate real-world adoption for frontier models. It also says the market may be under-weighting quality in agentic workflows where a wrong answer is expensive — we don't have the data yet to distinguish "agent routers are efficient" from "agent routers are cheap and the downstream users pay for the mistakes".
More rigorous join of benchmark scores × usage is on the roadmap — requires a clean model-ID match across OpenRouter permaslugs and our benchmark DB. Coming next.
Methodology & related pages
- Data source:
openrouter.ai/rankingsRSC stream. 53 weeks of vendor-level tokens, 77 weeks of model-level. Captured 2026-04-14. - Top-9 slice: OpenRouter charts the top-9 vendors each week and rolls the rest into "Others". Vendors that fall out of top-9 show 0 in weeks where they were lower than the 9th-place vendor.
- Chinese labs = Qwen, DeepSeek, MiniMax, Xiaomi, Z.ai (Zhipu), StepFun, MoonshotAI, TNG Tech. NVIDIA Nemotron and Microsoft Phi counted as Western.
- Snapshot leaderboard: /agentic/openrouter-apps · /agentic/openrouter-models · /agentic/openrouter-categories
Spotted a market trend we missed?
Seeing a model, vendor, or app shift that isn't reflected here? Tell us — we reply within 48 hours and update the analysis.