ToolScore
LLM APIs 4 min read

The 10 Most-Used Models on OpenRouter, Priced: The $0.006 Input Champion and the $20 Output Flagships

53.8T tokens/week — one model the number that matters here

By Stan K.

Live pull from OpenRouter's API (Nov 11, 2026): the ten models eating the world's token traffic — DeepSeek V4.1 Flash at 53.8T tokens/week leads — with input/cached/output per-1M prices, what each tier is for, and the Wafer anomaly that hosts GLM-5.3 20× cheaper than its own API.

Forget launch-day leaderboards. The token leaderboard is where the market votes with real money. We pulled OpenRouter’s weekly usage rankings and live price list on Nov 11, 2026 — this is what the world actually runs on.

The top 10 by weekly token volume

#ModelWeekly tokensTrend$/1M in$/1M cached$/1M out
1DeepSeek V4.1 Flash 101753.8T−2%$0.006$0.005$0.50
2Hy4 (Tencent)18.8Tnew———
3DeepSeek V4.1 Flash16.9T+8%$0.051$0.018$0.567
4GLM 5.3 Flash11.3T−31%$0.0399$0.03$0.0784
5GPT-6 Luna7.04T−24%$0.10$0.01$0.50
6Step 5 Preview6.94T+4%$1.00$0.10$2.70
7Claude Opus 5.56.76T+1%$4.00$0.20$20.00
8Crouton Alpha (Stealth)5.07T>999%———
9Kimi K3.1 Pro4.42T+46%$2.00$0.20$10.00
10MiniMax M3.1 Flash4.33T−33%$0.15$0.015$0.60

The pattern is blunt: volume flows to the cheap seats. The #1 and #3 models combined carry 70T tokens a week at input prices under six cents per million. The flagships — Opus 5.5, Kimi K3.1 Pro — carry a fraction of the volume at 30–700× the input price.

What each tier is actually for

The bulk fuel tier ($0.006–0.15 in). DeepSeek Flash 1017, GLM 5.3 Flash, GPT-6 Luna, MiniMax M3.1 Flash. This is where agents should do 90% of their work: reading files, writing boilerplate, running tool loops. At a 90/8/2 agent mix, 100M tokens of this tier costs $1.80–3.20. A billion tokens a month of Flash-grade fuel is a $20–30 API bill — the number that makes or breaks every coding-plan subscription math.

The smart tier ($1–4 in). Step 5, Kimi K3.1 Pro. Planning, architecture, tricky debugging at a third of flagship prices.

The flagship tier ($4–10 in, $20–50 out). Claude Opus 5.5 ($4/$0.20/$20), Sonnet 5.5 and Gemini 4 Argon/Carbon ($2/$0.10/$10), Qwen3.8 Max Prime ($4/$0.50/$12), GPT-6 Astra ($10/$1/$50). Worth it exactly when one-shot correctness beats retry economics — the same task that takes Opus one pass may take a Flash model three, and even then the Flash route is often cheaper. Our agent-quality rankings measure that trade-off.

Input price per 1M tokens vs weekly volume — the market votes cheap
DS Flash 1017 (53.8T/wk) $0.006
GLM 5.3 Flash (11.3T/wk) $0.0399
GPT-6 Luna (7T/wk) $0.10
Step 5 (6.9T/wk) $1.00
Kimi K3.1 Pro (4.4T/wk) $2.00
Claude Opus 5.5 (6.8T/wk) $4.00

The Wafer anomaly: same model, 20× cheaper input

One finding from the live pull deserves its own paragraph: GLM-5.3 through OpenRouter’s Wafer host runs $0.0699 input / $0.0689 cached / $1.40 output — while Z.ai’s own API lists $1.40 in / $0.26 cached / $4.40 out. Same model, ~20× cheaper input, 3× cheaper output. Cached tokens barely differ ($0.0689 vs $0.26), which tells you the gap is aimed at fresh-input-heavy workloads. Cache-heavy agent workloads close much of the gap on their own — one reason a measured GLM Pro month still crushed its plan price — but for fresh-input-heavy API users, the route you choose is worth more than the plan you choose.

Caveats that matter

Prices are a live snapshot (Nov 11, 2026) — OpenRouter hosts reprice weekly, and this market moves fast (DeepSeek’s August 2026 price rise, Z.ai’s July 2026 credit rework, Google’s October 2026 Code Assist sunset). The :free and :batch variants we excluded can halve costs again at the price of latency. And two of the top-10 models (Hy4, Crouton Alpha) don’t publish stable public pricing yet — they’re listed with volume but no verified per-1M rates.

For how these prices feed into subscription break-even math, see the plan-vs-API article and the full coding-plans ranking.

Numbers change — the math doesn't
Token prices and plan limits shift monthly; the structural arguments in this guide age slowly. For current per-tool numbers see the live rankings.