Codesota · Agentic · OpenRouter appsThe 30-day cost leaderboardSnapshot · 2026-06-23
§ Agentic · Market tracking

Who burns the most on
LLM inference.

OpenRouter publishes per-app token counts for every AI tool routing through it. We read the public app listings, join each model to live pricing, and rank the top 46 apps by estimated 30-day spend. Cost assumes a 72/28 input/output split — a reasonable approximation for tool-heavy agentic workloads.

§ 01 · Findings

Three things the data says.

  • Top line item
    Hermes Agent
    $17.74M/month — 23.3% of all spend across the top 46 apps. It moves 20.57T tokens at a blended $0.86/M.
  • Cheapest routing
    HammerAI
    Runs the leanest mix at $0.03/M blended — roughly 232.4× cheaper per token than Codex at $7.69/M. Same job, different model routing.
  • Tokens ≠ dollars
    Model mix dominates.
    Claude Code moves 2.80T (more than Hermes Agent's 20.57T) but spends $14.17M vs Hermes's $17.74M. Same category, different checkbook.
§ 02 · By cost

Apps ranked by monthly cost.

Each app's top-20 model mix, priced against OpenRouter's live catalog. A few models (stealth/alpha lanes, unlisted internal routes) carry no price and are excluded from the dollar total but still count toward the token total.

#AppCategoriesTokensRequestsMonthlyBlended $/M
1Hermes Agent
openrouter.ai/apps/hermes-agent
20.57T0$17.74M$0.86
2Claude Code
openrouter.ai/apps/claude-code
2.80T0$14.17M$5.06
3Descript
openrouter.ai/apps/descript
2.05T0$11.61M$5.68
4OpenClaw
openrouter.ai/apps/openclaw
4.04T0$8.97M$2.22
5Kilo Code
openrouter.ai/apps/kilo-code
5.84T0$4.77M$0.82
6pi
openrouter.ai/apps/pi
1.50T0$2.82M$1.88
7Codex
openrouter.ai/apps/codex
353.8B0$2.72M$7.69
8Cline
openrouter.ai/apps/cline
781.7B0$2.25M$2.88
9Lemonade
openrouter.ai/apps/lemonade
971.6B0$1.44M$1.48
10Pieces.app: Ambient Artificial Memory
openrouter.ai/apps/pieces
544.4B0$1.07M$1.96
11Ito
openrouter.ai/apps/ito
129.0B0$682K$5.29
12Stanley
openrouter.ai/apps/stan
148.9B0$677K$4.55
13SillyTavern
openrouter.ai/apps/sillytavern
220.1B0$655K$2.98
14Open WebUI
openrouter.ai/apps/open-webui
131.0B0$652K$4.98
15OpenHands
openrouter.ai/apps/openhands
219.5B0$623K$2.84
16Portkey AI
openrouter.ai/apps/portkey-ai
202.4B0$597K$2.95
17ISEKAI ZERO
openrouter.ai/apps/isekai-zero
907.6B0$578K$0.64
18Roo Code
openrouter.ai/apps/roo-code
254.8B0$561K$2.20
19LangChain
openrouter.ai/apps/langchain
189.7B0$440K$2.32
20Janitor AI
openrouter.ai/apps/janitor-ai
813.7B0$412K$0.51
21Agent Zero
openrouter.ai/apps/agent-zero
67.1B0$375K$5.59
22GitLawb
openrouter.ai/apps/gitlawb
612.4B0$337K$0.55
23Zed Editor
openrouter.ai/apps/zed-editor
162.4B0$245K$1.51
24Studs.gg
openrouter.ai/apps/studsgg
62.1B0$221K$3.56
25VidMuse
openrouter.ai/apps/vidmuse
31.3B0$198K$6.32
26Mira is the leading AI agent inside Telegram
openrouter.ai/apps/mira
503.2B0$195K$0.39
27GDevelop
openrouter.ai/apps/gdevelop
200.0B0$170K$0.85
28Chub AI
openrouter.ai/apps/chub-ai
102.9B0$166K$1.62
29novelcrafter
openrouter.ai/apps/novelcrafter
23.1B0$148K$6.42
30LibreChat
openrouter.ai/apps/librechat
29.5B0$125K$4.25
31Nous Research API
openrouter.ai/apps/nous-research-api
118.8B0$91K$0.77
32nanobot
openrouter.ai/apps/nanobot
36.3B0$72K$1.99
33Craft
openrouter.ai/apps/craft-rpgs
145.6B0$57K$0.39
34Qwen Code
openrouter.ai/apps/qwen-code
98.6B0$56K$0.57
35extra.email
openrouter.ai/apps/extra-email
172.2B0$27K$0.16
36Aider
openrouter.ai/apps/aider
5.8B0$19K$3.27
37ZeroClaw
openrouter.ai/apps/zeroclaw
9.5B0$13K$1.33
38openbench
openrouter.ai/apps/openbench
11.5B0$8K$0.71
39HammerAI
openrouter.ai/apps/hammerai
222.1B0$7K$0.03
40Gobii Meta Gobii Skill Eval
openrouter.ai/apps/gobii
6.4B0$6K$1.00
41TeleClaw
openrouter.ai/apps/teleclaw
36.3B0$4K$0.12
42Clark - Personal Agent
openrouter.ai/apps/clark
8.2B0$3K$0.34
43MinutaIA
openrouter.ai/apps/minuta-ia
3.6B0$3K$0.77
44CoffeeCat AI Image Generator
openrouter.ai/apps/coffeecat-ai
18.8B0$2K$0.13
45Oh My Pi
openrouter.ai/apps/oh-my-pi
3.0B0$1K$0.45
46Fish Audio - the most realistic AI voice generator
openrouter.ai/apps/fish-audio
2.0B0$1K$0.52
Fig 1 · Monthly cost leaderboard. Copper rows are the top three line-items; unpriced stealth/alpha lanes drop from the dollar total but still count toward tokens.
§ 03 · By tokens

Apps ranked by raw token volume.

Full tracked app list including tail entries without model breakdowns. Volume-only ranking — no cost join.

#AppGlobal rankTokensRequestsVolume curve
1Hermes Agent#120.57T0
2Kilo Code#25.84T0
3OpenClaw#34.04T0
4Claude Code#42.80T0
5Descript#52.05T0
6pi#61.50T0
7Lemonade#7971.6B0
8ISEKAI ZERO#8907.6B0
9Janitor AI#9813.7B0
10Cline#10781.7B0
11GitLawb#11612.4B0
12Pieces.app: Ambient Artificial Memory#12544.4B0
13Mira is the leading AI agent inside Telegram#13503.2B0
14Codex#14353.8B0
15Roo Code#15254.8B0
16HammerAI#16222.1B0
17SillyTavern#17220.1B0
18OpenHands#18219.5B0
19Portkey AI#19202.4B0
20GDevelop#20200.0B0
21LangChain#21189.7B0
22extra.email#22172.2B0
23Zed Editor#23162.4B0
24Stanley#24148.9B0
25Craft#25145.6B0
26Open WebUI#26131.0B0
27Ito#27129.0B0
28Nous Research API#28118.8B0
29Chub AI#29102.9B0
30Qwen Code#3098.6B0
31Agent Zero#3167.1B0
32Studs.gg#3262.1B0
33TeleClaw#3336.3B0
34nanobot#3436.3B0
35VidMuse#3531.3B0
36LibreChat#3629.5B0
37novelcrafter#3723.1B0
38CoffeeCat AI Image Generator#3818.8B0
39openbench#3911.5B0
40ZeroClaw#409.5B0
41Clark - Personal Agent#418.2B0
42Gobii Meta Gobii Skill Eval#426.4B0
43Aider#435.8B0
44MinutaIA#443.6B0
45Oh My Pi#453.0B0
46Fish Audio - the most realistic AI voice generator#462.0B0
Fig 2 · All tracked apps ranked by tokens. The sparkline on row 1 shows the top-10 volume curve — long-tailed, as expected from a power-law market.
§ 04 · Methodology

Where these numbers come from.

  • Data source.
    openrouter.ai/apps and each app's detail page, fetched via the React Server Components stream (RSC: 1 header). Captured 2026-06-23.
  • Pricing.
    Public openrouter.ai/api/v1/models catalog. Permaslugs (e.g. anthropic/claude-4.6-opus-20260205) are matched back to base catalog IDs after stripping date and preview suffixes.
  • Blended $/M.
    price_in × 0.72 + price_out × 0.28. Agentic workloads are context-heavy because every tool round replays the conversation. A 50/50 split inflates dollar totals by ~10–15%.
  • Top-20 models per app.
    OpenRouter surfaces the top 20 models by tokens for each app. Apps may use more models beyond that slice at long-tail volume, so reported spend is a lower bound.
  • Unpriced models.
    Stealth/alpha lanes (e.g. openrouter/hunter-alpha) and unlisted providers are dropped from the dollar total. Free tiers are priced at $0.
  • Not a quality ranking.
    This page measures compute spend, not output quality. For task performance see our agentic benchmarks.
§ Final · Related

What to read next.

We reply within 48 hours

App missing from the leaderboard?

Know an app routing serious volume through OpenRouter that isn't here? Send the slug — we'll include it in the next refresh.

Tell us what you found →
✓ No newsletter✓ Real humans read this✓ 30 seconds to send