All benchmarks

Chinese labs serve 57% of OpenRouter's tokens and earn 25% of its dollars

OpenAI, Anthropic and Google take 67% of the dollars on 20% of the tokens. These are OpenRouter's own billed figures for Sep 14 – Oct 11, 2026, the weeks with spend for every model.

Data from openrouter.ai and artificialanalysis.ai · through Oct 18, 2026

Oct 20 '25 – Oct 18 '26
Billed on OpenRouter, Oct 20, 2025 – Oct 18, 2026
$599M
2,267T tokens · 154B requests
$11.5M a week on average · 68% attributed to a named model
Average week, last 4 vs first 4 weeks shown
7.2x the dollars
on 29x the tokens and 8.2x the requests
Anthropic, Sep 14 – Oct 11, 2026
30% of dollars
on 4.6% of tokens
Average request, last 4 vs first 4 weeks shown
21K tokens
up from 5.8K. Cost per request $0.0027 → $0.0024
Intelligence Index 20+, vs a year ago
22x cheaper
Chapter 1

The money

Billed dollars by week, lab and model. Follows the range above.

Average weekly spend rose from $2.58M to $18.7M

Billed dollars per week by lab, Oct 20, 2025 – Oct 18, 2026. The lab colors jump at Sep 14 because per-model dollars start that week, not because spending shifted. Tap or hover a week for its split.

$5M
$10M
$15M
$20M
$25M
$30M
Nov 10
Jan 5
Mar 2
Apr 27
Jun 22
Aug 17
Oct 12
Week of Oct 12$320K
Unattributed$151K47%
Other labs$14K4.5%
Chinese labs$31K9.7%
OpenAI$61K19%
Anthropic$62K19%
164.0T tokens · 8.6B requests · $0.002 per M tokens
Anthropic$239MOpenAI$65.2MGoogle$43.1MChinese labs$57.1MOther labs$3.72MUnattributed$190M

Unattributed is spend OpenRouter reports only as “Others”. Before Sep 14, only each week's top 9 models by spend are named, so lab totals are floors. Legend totals cover the 52 weeks shown.

A token got 4.0x cheaper; a request cost about the same

1x is the weekly average for Oct 20 – Nov 16, 2025. Over Sep 21 – Oct 18, 2026, spend ran at 7.2x, requests at 8.2x and tokens at 29x, so each request carried 3.6x the tokens. Log scale once the lines spread past 3x.

1x10xNov 10Jan 5Mar 2Apr 27Jun 22Aug 17Oct 12Tokens 29xRequests 8.2xSpend 7.2x
Billed dollars per request
Average $0.0027 → $0.0024
$0$0.0020$0.0040$0.0060$0.0080Nov 10Jan 5Mar 2Apr 27Jun 22Aug 17Oct 12
Tokens per request
Average 5.8K → 21K
05K10K15K20K25KNov 10Jan 5Mar 2Apr 27Jun 22Aug 17Oct 12

Anthropic takes the most dollars; Chinese labs serve the most tokens

Sep 14 – Oct 11, 2026 · the 4 weeks with dollars for every model, for any range

Share of tokens (top) vs share of dollars (bottom). A band that widens toward the bottom charges more per token than average.

Tokens
4.6%30%11%26%4.3%11%57%25%20%4.4%
Dollars

Anthropic: 4.6% of tokens, 30% of dollars ($29.7M, $1.06 per million tokens)

LabTokensDollars$ / M tok
Anthropic4.6%30%$1.06
OpenAI11%26%$0.37
Google4.3%11%$0.43
Z.ai12%9.5%$0.13
DeepSeek25%8.1%$0.052
Moonshot AI1.2%4.9%$0.67
xAI0.4%2.3%$0.83
Qwen1.2%1.3%$0.18
Meta1.5%0.9%$0.094
Xiaomi6.9%0.4%$0.010
Tencent8.0%0.4%$0.009
TypeSafe1.6%0.4%$0.038

Top 12 of 25 labs by dollars.

Half the dollars took 3 models in the week of Oct 20, 9 by Oct 12

The biggest model's share fell from 27% to 13%. Built from each week's published top 9 by spend, so “more than 9” is as far as it can count.

Models holding half the dollars
more than 902468Nov 10Jan 5Mar 2Apr 27Jun 22Aug 17Oct 12
Share of the week's dollars
0%20%40%60%80%100%Nov 10Jan 5Mar 2Apr 27Jun 22Aug 17Oct 12Top 9 53%Top 1 13%

With every model counted, Oct 5 – Oct 11, 2026 (1 week): 9 models held half the dollars and the top 10 held 54% (HHI 473 out of 10,000).

Claude Opus 4.7 earned the most, Oct 20, 2025 – Oct 18, 2026

Billed dollars per model. The trend line shows weekly dollars over the 52 weeks shown.

  1. 1Claude Opus 4.7$43.7M 7.3%$1.71/M
  2. 2Claude Sonnet 4.6$40.3M 6.7%$1.13/M
  3. 3Claude Opus 4.8$31.0M 5.2%$1.70/M
  4. 4Claude Opus 4.6$30.4M 5.1%$2.10/M
  5. 5Claude Opus 5$21.7M 3.6%$1.49/M
  6. 6Claude Sonnet 4.5$21.4M 3.6%$1.49/M
  7. 7GPT-5.6 Sol$15.4M 2.6%$0.86/M
  8. 8GPT-5.5$14.8M 2.5%$1.76/M
  9. 9GLM 5.2open$13.9M 2.3%$0.35/M
  10. 10Kimi K3open$13.7M 2.3%$0.65/M
  11. 11GPT-6 Astranew$12.5M 2.1%$2.13/M
  12. 12Claude Opus 4.5$11.0M 1.8%$2.62/M

Bar color is the lab group; the percentage is the model's share of all billed dollars in the range. Gaps in a trend are weeks the model wasn't named. Weeks before full leaderboards name only the top 9, so these totals are floors.

Chapter 2

What the money buys

Topics, tasks, apps and the shape of a request. Most of these are 30-day snapshots; each section names its window.

Software & technology is 55% of the spend, from 30% of requests but 65% of tokens

Last 30 days · Oct 23 – Nov 21, 2026

Share of spend, tokens and requests by topic and by task, as OpenRouter classifies prompts. Log scale, so the gap between dots is the ratio. Price is what a category pays per token vs the average. Tap or hover a row for its top models.

SpendRequests (topics only)Tokens
By topic
Last 30 days · Oct 23 – Nov 21, 2026
2%5%10%20%50%Spend
Software & technology55%
General8.3%
Arts & entertainment7.7%
Business operations5.3%
Science & engineering3.9%
Gaming3.3%
Finance2.9%
Education2.1%
Software & technology
Spend55%
Requests30%
Tokens65%
0.8x the average price per token · ≈$0 est.
Top models by spend, change vs prior 30 days
Claude Opus 5.519%+1.0pp
Kimi K3.1 Pro7.2%+7.2pp
GPT-6.1 Sol6.2%+2.5pp
By task
Last 30 days · Oct 23 – Nov 21, 2026
0.1%1%10%Spend
Software40%
Software.implementation16%
Software.debugging7.4%
Software.code review5.5%
Data20%
Data.classification tagging9.1%
Data.extraction5.3%
Data.evaluation2.1%
Information15%
Information.context grounded qa6.7%
Information.research synthesis3.2%
Information.planning decision support1.8%
Writing6.8%
Writing.drafting4.1%
Writing.rewriting editing1.4%
Writing.summarization0.7%
Assistance6.5%
Assistance.service issue resolution2.0%
Assistance.transactional assistance1.9%
Assistance.workflow control1.7%
Creative6.2%
Creative.interactive roleplay5.1%
Creative.creative writing1.1%
Media4.3%
Media.image generation editing2.6%
Media.video generation1.2%
Media.speech transcription0.3%
Retrieval0.3%
Retrieval.embedding0.1%
Retrieval.reranking0.1%
Software
Spend40%
Tokens50%
0.8x the average price per token · ≈$0 est.
Spend share by task group, weekly snapshots of the last 30 days
Jun 22 – Nov 16, 2026
Software
40%
+0.0pp since Jun 22
Data
20%
+13pp since Jun 22
Information
15%
+0.0pp since Jun 22
Writing
6.8%
+0.0pp since Jun 22
Assistance
6.5%
+0.0pp since Jun 22
Creative
6.2%
+0.0pp since Jun 22
Media
4.3%
+0.0pp since Jun 22
Retrieval
0.3%
+0.0pp since Jun 22

The model moves a session's cost up to 92x; the app, a median 1.5x

Last 30 days · Oct 3 – Nov 1, 2026

Median cost of one coding-agent session. Each row is a model, each mark an app, log scale. Tap or hover a row for every app's cost.

Same model, different app
1.5x
median gap between a model's priciest and cheapest app, across 20 models in 3+ apps
Same app, different model
up to 92x
90th vs 10th percentile model cost in one app, 10–49 turns
Hermes Agent31x
Claude Code77x
Kilo Code29x
Codex92x
Sessions of 50+ turns
6.3x
the median cost of a 10–49 turns session, same model and app
$0.10$0.20$0.50$1$2Apps
Claude Fable 5.11.1x
GPT-6 Astra1.1x
Claude Opus 51.0x
Claude Opus 4.81.4x
Claude Opus 5.51.8x
GPT-5.5
Gemini 4 Argon1.6x
Claude Opus 4.61.7x
Grok 4.71.2x
GPT-5.6 Sol1.1x
Claude Sonnet 4.61.2x
Kimi K3.1 Pro1.8x
Claude Sonnet 4.51.3x
Qwen3.8 Max (0902)1.4x
Claude Sonnet 51.0x
Kimi K31.7x
Claude Sonnet 5.51.4x
GPT-6 Sol1.2x
GPT-6.1 Sol1.2x
GPT-5.6 Terra1.1x
Kimi K2.7 Code1.1x
Mistral Large 4
Gemini 3.8 Flash1.7x
DeepSeek V4 Pro 08131.2x
Hy4 preview1.3x
DeepSeek V4 Pro 04231.1x
Claude Haiku 4.51.3x
GLM 5.21.2x
GLM 5.31.6x
MiniMax M31.5x
Claude Fable 5.1
Median cost per session, 10–49 turns
Hermes Agent$3.20
Claude Code$3.65
Kilo Code–
Codex–
1.1x between apps
Top apps by tokens
Last 30 days · Oct 23 – Nov 21, 2026
  1. 1Hermes Agent55.8T
  2. 2Claude Code45.8T
  3. 3Kilo Code25.9T
  4. 4pi20.6T
  5. 5Codex15.1T
  6. 6OpenClaw13.8T
  7. 7omp10.2T
  8. 8Strix9.52T
  9. 9DeepSeek Harness9.43T
  10. 10Portkey AI7.51T
Top apps by estimated spend
Last 30 days · Oct 23 – Nov 21, 2026
  1. 1Portkey AI≈$0
  2. 2pi≈$0
  3. 3Claude Code≈$0
  4. 4Hermes Agent≈$0
  5. 5omp≈$0
  6. 6Nous Research API≈$0
  7. 7OpenHands≈$0
  8. 8Vogel AI≈$0
  9. 9Command Code≈$0
  10. 10Kilo Code≈$0

Estimate: each app's tokens on 25 tracked models × what those models bill through their top apps. Those apps carry 0.0% of the models' billed spend, so this is a floor. Coding and agent apps take 0.0% of it.

Prompts over 100K tokens grew from 1.5% to 6.5% of requests

Apr 20 – Nov 15, 2026 · 30 weeks

Share of requests by prompt length, by week. Long prompts carry files, tool results and conversation history.

0%25%50%75%100%May 18Jun 22Jul 27Aug 31Oct 5Nov 9< 1K tokens 30%1K–10K tokens 46%10K–100K tokens 18%100K–1M tokens 6.5%1M–10M tokens 0.0%
< 1K tokens30%1K–10K tokens46%10K–100K tokens18%100K–1M tokens6.5%1M–10M tokens0.0%
Tool calls per request
Jul 20 – Nov 15, 2026 · 0.18 → 0.26
0.000.100.200.30Jul 27Aug 17Sep 7Sep 28Oct 19Nov 9
Where the output goes
Last 30 days · Oct 23 – Nov 21, 2026
Completion tokens spent reasoning
58%
thinking before the answer, billed as output

Across 25 tracked models.

Who pays for the tokens
Last 30 days · Oct 23 – Nov 21, 2026
Paid through OpenRouter0.0%Bring your own key28%Free models72%

Share of 971.4T tokens ($0 billed). Bring your own key: the user pays the provider, OpenRouter bills only its fee. Free: models billed under $0.01 per million tokens.

Chapter 3

Sticker vs street

List prices against what OpenRouter actually billed, and what one finished task costs.

Cache reads, last 30 days
91%
of prompt tokens. Billed 0.0% of what the same tokens cost at list price without caching
What $1 buys vs Oct 13, 2025
4.3x the tokens
10.4x weighted by score (average 16 → 39). Weekly averages, Oct 13 – Nov 9, 2025 vs Sep 21 – Oct 18, 2026

A billed token costs 14% of its list-price blend

Sep 14 – Oct 11, 2026 (4 weeks)

Per million tokens: Artificial Analysis's list price (3 input tokens to 1 output), its cached-agent blend (7 cache reads, 2 input, 1 output) and what OpenRouter billed. Top 12 scored models by dollars. Log scale.

List priceCached-agent blendBilled
GPT-6 Astra
11%
Claude Opus 5.5
12%
GPT-5.6 Sol
9.6%
Claude Opus 5
14%
Kimi K3
12%
DeepSeek V4.1 Flash
7.0%
GLM 5.3
14%
Claude Fable 5.1
8.9%
Claude Sonnet 5
16%
GLM 5.3 Flash
24%
Gemini 3.8 Flash
22%
GLM 5.2
18%

Right column: billed as a share of list. Across 90 models, billed dollars were 14% of list and 36% of the cached blend. Most of the gap is the traffic itself: about 45 input tokens per output token, with 91% of prompt tokens read from cache (last 30 days). The median model billed 79% of what Artificial Analysis paid per token in its own test runs.

Claude Opus 5.5 is cheaper per token than GPT-6 Astra, but 4.4x pricier per task

Sep 14 – Oct 11, 2026 (4 weeks)

Left: models ranked by billed price per million tokens. Right: ranked by what one Artificial Analysis index task costs at that price, which counts the tokens each model spends thinking. Cheapest on top.

$ / M tokens$ / taskGPT-6 Astra $2.19$1.65 GPT-6 AstraClaude Opus 5.5 $0.94$7.25 Claude Opus 5.5Kimi K3 $0.69$1.03 Kimi K3DeepSeek V4.1 Flash $0.037$0.20 DeepSeek V4.1 FlashGLM 5.3 $0.30$1.34 GLM 5.3Claude Fable 5.1 $1.79$6.04 Claude Fable 5.1GLM 5.3 Flash $0.057$0.35 GLM 5.3 FlashGemini 3.8 Flash $0.33$1.57 Gemini 3.8 FlashClaude Sonnet 5.5 $0.55$7.86 Claude Sonnet 5.5GPT-6.1 Sol $0.44$0.43 GPT-6.1 SolGrok 4.7 $0.79$2.62 Grok 4.7GPT-6 Luna $0.045$0.083 GPT-6 LunaMuse Spark 1.3 $0.42$1.90 Muse Spark 1.3Qwen3.8 Max (0902) $0.51$5.66 Qwen3.8 Max (0902)Qwen3.8 27B $0.22$0.71 Qwen3.8 27BMiMo-V2.6-Pro $0.052$0.15 MiMo-V2.6-ProMiniMax M3 $0.032$0.16 MiniMax M3Claude Haiku 5.5 $0.099$0.52 Claude Haiku 5.5Step 5 Preview $0.001$0.0053 Step 5 Preview

Claude Opus 5.5: $0.94 per million tokens, 119K tokens per task, $7.25 per task, score 57.6

A cheaper equal existed for 68% of dollars

That share went to a model while another scoring at least as high cost a third as much per token. Judged per task, it is 49%. The score ignores speed, context length and tool reliability, which also drive the choice. Covers 85% of the window's dollars.

Dollars per task per index point, by lab
Anthropic$0.11
Qwen$0.067
xAI$0.053
Moonshot AI$0.045
OpenAI$0.044
Google$0.038
Z.ai$0.032
DeepSeek$0.0073

Weighted by each lab's dollars, at Artificial Analysis list prices. Lower is cheaper intelligence.

Anthropic serves 0.0% of Claude spend, Google Vertex 0.0%

Last 30 days · Oct 23 – Nov 21, 2026

Which provider served the tokens behind each model. Estimated from OpenRouter's sampled provider mix for each model, weighted by price. These are sampled shares, not billed dollars.

Who serves Claude models, est. share of spend
Anthropic0.0%Google Vertex0.0%Azure0.0%Amazon Bedrock0.0%
All tracked models: est. share of spend vs sampled share of tokens
SpendTokens
0.1%0.2%0.5%1%2%Spend
Nebius Token Factory0.0%
Wafer0.0%
inference.net0.0%
DigitalOcean0.0%
River0.0%
Relace0.0%
Moonshot AI0.0%
Parasail0.0%
Morph0.0%
Decart0.0%
Fireworks0.0%
Makora0.0%
Nebius Token Factory
Spend (est.)0.0%
Tokens (sampled)1.5%
0.0x the average price per token

25 tracked models. Log scale. Price is spend share ÷ token share.

Chapter 4

The frontier

How far spending trails the best-scoring model, how fast new models take over, and what each score level costs.

Median model age per dollar, Sep 14 – Oct 11
53 days
Half of all dollars went to models younger than that. Per token, 28 days
Models under 30 days old, same weeks
36% of dollars
and 53% of tokens
Models over a year old, same weeks
3.5% of dollars
while 69% went to models under 90 days old

The average dollar trails the frontier by 1 month, the average token by 5 months

The highest Artificial Analysis score on offer each week, against the score of the average dollar and the average token, Oct 20, 2025 – Oct 18, 2026. The lag, for the week of Oct 12, is how long ago the frontier stood where the average stands now. Built from each week's top 9 models; 49% of that week's dollars are on scored models.

0204060Nov 10Jan 5Mar 2Apr 27Jun 22Aug 17Oct 12Frontier 57.6Avg dollar 51.2Avg token 40.7
Share of dollars on models within 5 points of the top score

24% on average while OpenAI held the top score (18 weeks shown) · 23% on average while Anthropic held the top score (33 weeks shown)

0%10%20%30%40%50%Nov 10Jan 5Mar 2Apr 27Jun 22Aug 17Oct 12

Two weeks after a successor launches, the predecessor keeps a median 55% of its dollar share

Each line is one model, aligned on its launch week, across all history (not the range). Gaps are weeks it fell outside the published top 9 by spend (the tooltip gives the cutoff) or weeks before this data starts.

Claude Opus
Share of each week's dollars by weeks since launch, from each week's published top 9
0%10%20%30%launch+3w+6w+9w+12w+15wOpus 5.5Opus 4.7Opus 5Opus 4.8
Predecessor shareWeek before+2 wk+4 wk
Opus 4.7 → Opus 4.826%23%21%
Opus 4.8 → Opus 519%4.8%3.8%
Opus 5 → Opus 5.56.8%4.1%–
Claude Sonnet
Share of each week's dollars by weeks since launch, from each week's published top 9
0%10%20%30%40%launch+3w+6w+9w+12w+15wSonnet 4.5Sonnet 4.6Sonnet 5.5Sonnet 5
Predecessor shareWeek before+2 wk+4 wk
Sonnet 4.5 → Sonnet 4.615%7.8%6.1%
Sonnet 4.6 → Sonnet 58.8%4.9%4.9%
GPT flagship
Share of each week's dollars by weeks since launch, from each week's published top 9
0%5%10%15%20%launch+3w+6w+9w+12w+15wGPT-6 AstraGPT-5.6 SolGPT-5.5GPT-5.4
Predecessor shareWeek before+2 wk+4 wk
GPT-5.4 → GPT-5.55.3%3.3%3.4%
GPT-5.5 → GPT-5.6 Sol9.6%3.6%–
GPT-5.6 Sol → GPT-6 Astra6.6%6.3%4.3%
Gemini Pro
Share of each week's dollars by weeks since launch, from each week's published top 9
0%2%4%6%8%10%launch+3w+6w+9w+12w+15w3.1 Pro Preview3 Pro Preview
Predecessor shareWeek before+2 wk+4 wk
2.5 Pro → 3 Pro Preview13%8.3%6.6%
3 Pro Preview → 3.1 Pro Preview5.7%3.1%–

A 20+ model got 22x cheaper in a year

Oct '24 – Nov '26, monthly

Each line is the cheapest model scoring at least that much on the Artificial Analysis Intelligence Index, by month. A line steps up when the cheapest model gets pricier or is delisted. Log scale.

$.01$.1$1$10$100Nov '24May '25Nov '25May '26Nov '2650+ $4.0030+ 40+ $0.05010+ 20+ $0.032

The top score is Claude Opus 5.5's 57.6, up 46 points in 24 months from o1-preview's 11.4. No model reaches 60 on this version of the index.

ScoreYear agoNowCheaper
10+$0.080$0.0322.5x
20+$0.69$0.03222x
30+–$0.050–
40+–$0.050–
50+–$4.00–

Price per million tokens, 3 parts input to 1 part output. Scores first reached within the last year show a dash. “Halves every” is the trend since the tier was first reached.

Every model: dollar share, billed vs list price, score and cost per task, Sep 14 – Oct 11, 2026 (4 weeks)
Model$ shareToken shareBilled $/MList $/MScore$ / taskTokens / taskWeightsLaunched
GPT-6 Astra11%0.8%$2.19$20.052.7$1.6527Kclosed2026-09-03
Claude Opus 5.58.1%1.4%$0.94$8.0057.6$7.25119Kclosed2026-09-21
GPT-5.6 Sol5.1%1.1%$0.77$8.0047.0$1.99*–closed2026-07-09
Claude Opus 55.1%0.6%$1.40$10.050.8$5.86*–closed2026-07-23
Kimi K34.6%1.1%$0.69$6.0043.6$1.0348Kopen2026-07-15
DeepSeek V4.1 Flash3.8%17%$0.037$0.5339.5$0.2089Kopen2026-09-10
GLM 5.33.8%2.1%$0.30$2.1544.8$1.3471Kopen2026-08-16
Claude Fable 5.13.3%0.3%$1.79$20.053.4$6.0478Kclosed2026-08-31
Claude Sonnet 53.1%0.8%$0.64$4.0038.2$5.09*–closed2026-06-30
GLM 5.3 Flash3.0%8.6%$0.057$0.2441.8$0.3569Kopen2026-08-26
Gemini 3.8 Flash2.9%1.4%$0.33$1.5040.9$1.5771Kclosed2026-09-02
GLM 5.22.4%1.0%$0.40$2.1533.7$1.47*–open2026-06-16
Claude Sonnet 4.62.2%0.2%$1.62$6.0030.1$2.49*–closed2026-02-17
Gemini 3.1 Pro Preview2.2%0.1%$2.42$4.5029.7$1.30*–closed2026-02-19
Claude Opus 4.82.1%0.2%$1.82$10.041.8$4.08*–closed2026-05-28
Claude Sonnet 5.51.8%0.5%$0.55$4.0056.0$7.86197Kclosed2026-09-28
DeepSeek V4 Flash 07311.4%4.6%$0.048$0.6634.3$0.22*–open2026-07-31
Gemini 3 Flash Preview1.3%0.5%$0.40$1.1326.3––closed2025-12-17
GPT-5.6 Luna1.3%4.0%$0.052$0.4537.3$0.18*–closed2026-07-09
GPT-6.1 Sol1.3%0.5%$0.44$4.0051.8$0.4338Kclosed2026-09-29
GPT-5.6 Terra1.3%0.3%$0.74$4.5042.1$1.40*–closed2026-07-09
GPT-6 Sol1.1%0.4%$0.41$4.0047.6$1.04*–closed2026-09-22
Grok 4.71.1%0.2%$0.79$3.0046.5$2.6281Kclosed2026-09-16
Gemini 3.7 Flash1.0%0.3%$0.53$1.5039.1$0.93*–closed2026-08-13
GPT-6 Astra Pro1.0%0.0%$5.99––––closed2026-09-03
DeepSeek V4 Pro 04231.0%0.5%$0.31$0.5430.4$0.12*–open2026-04-23
Claude Opus 4.61.0%0.1%$2.35$10.031.9––closed2026-02-05
Claude Sonnet 4.50.9%0.1%$2.09$6.0020.7$0.56*–closed2025-09-29
DeepSeek V4 Flash 04230.8%2.2%$0.060$0.1424.4$0.10*–open2026-04-23
DeepSeek V4 Pro 08130.8%0.7%$0.19$1.9836.0$0.67*–open2026-08-13
GPT-6 Luna0.7%2.6%$0.045$0.2038.1$0.08350Kclosed2026-09-22
Claude Fable 50.7%0.0%$3.25$20.049.6$8.75*–closed2026-06-09
Grok 4.60.7%0.1%$0.80$3.0044.3$1.49*–closed2026-08-10
Claude Haiku 4.50.7%0.2%$0.66$2.0016.9$0.28*–closed2025-10-01
Gemini 2.5 Flash0.6%0.3%$0.33$0.8513.1––closed2025-06-17
Muse Spark 1.30.6%0.3%$0.42$2.0048.1$1.9060Kclosed2026-09-02
Qwen3.8 Max (0902)0.6%0.2%$0.51$3.0045.4$5.66108Kclosed2026-09-02
GPT-5.50.5%0.0%$2.27$11.338.4$2.63*–closed2026-04-23
Gemini 3.5 Flash0.5%0.1%$1.31$3.3832.6$3.69*–closed2026-05-19
Gemini 2.5 Pro0.4%0.0%$1.81$3.4416.1$0.33*–closed2025-06-17
Gemini 3.1 Flash Lite0.4%0.3%$0.21––––closed2026-05-07
Hy4 preview0.4%6.0%$0.011––––open2026-08-27
Claude Opus 4.70.4%0.0%$2.18$10.040.7––closed2026-04-16
GPT-5.6 Sol Pro0.4%0.0%$2.24––––closed2026-07-09
Jev 1.130.4%1.6%$0.038––––closed2026-09-17
Gemini 3.6 Flash0.4%0.1%$0.73$1.5034.0$1.60*–closed2026-07-21
GPT-5.40.4%0.0%$1.43$5.6339.0––closed2026-03-05
Qwen3.8 27B0.3%0.3%$0.22$1.1333.7$0.7167Kopen2026-08-14
Gemma 4 31B0.3%0.3%$0.18$0.00014.7––open2026-04-02
Nano Banana 2 (Gemini 3.1 Flash Image)0.3%0.0%$21.6––––closed2026-05-28
Kimi K2.60.3%0.1%$0.43$1.7127.0$0.80*–open2026-04-20
GPT-4.10.3%0.0%$1.40$3.5012.7––closed2025-04-14
DeepSeek V3.20.3%0.2%$0.19$0.3221.5––open2025-12-01
Gemini 2.5 Flash Lite0.3%0.4%$0.10$0.178.6––closed2025-07-22
Grok 4.50.3%0.0%$1.03$3.0038.8$0.88*–closed2026-07-08
gpt-oss-120b0.2%0.4%$0.10$0.2611.6$0.11*–open2025-08-05
Muse Spark 1.3 Contributor0.2%1.2%$0.030––––closed2026-09-02
Gemini 3.5 Flash Lite0.2%0.2%$0.22$0.8522.2$0.19*–closed2026-07-21
GPT-5.20.2%0.0%$1.42$4.8130.4––closed2025-12-11
Claude Opus 4.50.2%0.0%$4.22$10.029.1––closed2025-11-24
Seedance 2.50.2%0.0%–––––closed2026-08-07
Grok 4.30.2%0.0%$1.05$1.5624.9$0.20*–closed2026-04-30
MiMo-V2.6-Pro0.2%0.6%$0.052$0.5446.3$0.1564Kopen2026-09-21
GPT-5 Mini0.2%0.1%$0.34$0.6916.8$0.053*–closed2025-08-07
GPT-5.4 Mini0.2%0.1%$0.41$1.6924.1$0.45*–closed2026-03-17
MiniMax M30.2%0.9%$0.032$0.5329.2$0.1648Kopen2026-05-31
GPT Image 2.5 Sunburst0.2%0.0%$13.5––––closed2026-09-08
Nano Banana Pro (Gemini 3 Pro Image)0.2%0.0%$41.5––––closed2026-05-28
GPT-4o-mini0.1%0.1%$0.16$0.266.7––closed2024-07-18
GLM 5.10.1%0.0%$0.86$1.9826.1$1.02*–open2026-04-06
Nano Banana Pro (Gemini 3 Pro Image Preview)0.1%0.0%$49.1––––closed2025-11-20
Claude Sonnet 40.1%0.0%$2.32–18.9––closed2025-05-22
Qwen3.5 397B A17B0.1%0.0%$0.69$1.3518.4$0.47*–open2026-02-16
Gemma 4 26B A4B 0.1%0.2%$0.085$0.1716.7––open2026-04-03
Claude Haiku 5.50.1%0.2%$0.099$0.2043.4$0.52162Kclosed2026-10-07
GPT-4.1 Mini0.1%0.1%$0.29$0.7010.2––closed2025-04-14
Qwen3.8 2.4T A95B0.1%0.0%$0.68$3.0040.0$2.15*–open2026-08-12
Grok 4.200.1%0.0%$0.88$3.0025.2––closed2026-03-09
GPT-4o0.1%0.0%$2.20–9.0––closed2024-05-13
Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)0.1%0.0%$12.5––––closed2026-06-30
MiMo-V2.6-Flash0.1%4.3%$0.004$0.1737.9$0.062*–open2026-09-21
Qwen3.8 Flash0.1%0.3%$0.052$0.2339.8$0.37*–open2026-08-26
GPT-5.6 Luna Pro0.1%0.1%$0.13––––closed2026-07-09
Gemini 3.1 Flash Lite Preview0.1%0.1%$0.25$0.5615.6$0.060*–closed2026-03-03
GPT-5.3-Codex0.1%0.0%$0.44$4.8132.5––closed2026-02-24
GLM 5.3 Prime0.1%0.0%$0.68––––closed2026-09-21
GPT-5.4 Image 20.1%0.0%$18.7––––closed2026-04-21
MiMo-V2.5-Pro0.1%0.1%$0.10$0.5426.0$0.054*–open2026-04-22
GLM 5.3 FlashX0.1%0.1%$0.12––––closed2026-09-18
GPT-6 Luna Pro0.1%0.2%$0.065––––closed2026-09-22
MiMo-V2.50.1%1.9%$0.005$0.1721.8$0.019*–open2026-04-22
Qwen3.7 Flash0.1%0.3%$0.035––––closed2026-07-27
gpt-oss-20b0.1%0.1%$0.055$0.0989.0$0.012*–open2025-08-05
Hy30.0%1.9%$0.004$0.2425.3$0.072*–open2026-07-06
GPT-5 Nano0.0%0.1%$0.051$0.1413.0––closed2025-08-07
Mistral Nemo0.0%0.2%$0.018––––open2024-07-19
Step 5 Preview0.0%1.4%$0.001$1.4343.7$0.005364Kopen2026-10-08
Solar Pro 40.0%0.8%$0.003$0.5328.1––closed2026-08-10
Qwen3 Embedding 8B0.0%0.1%$0.014––––open2025-10-28
Text Embedding 3 Small0.0%0.1%$0.018––––closed2025-10-30
Space Bunny Alpha0.0%9.3%$0.000––––closed2026-09-23
Nemotron 3 Ultra (free)0.0%3.7%–$1.0722.9$0.60*28Kopen2026-06-04
Laguna S 2.1 (free)0.0%0.7%–––––open2026-07-20
Nemotron 3.5 Lightning (free)0.0%0.4%–$0.09512.9$0.093*–open2026-08-07
Ling 3.0 Flash Fin (free)0.0%0.4%–$0.1122.6––closed2026-08-27
Dots3-Note Preview (free)0.0%0.4%–––––open2026-08-13
Nemotron 3 Super (free)0.0%0.3%–––––open2023-03-11
Solar Mini 40.0%0.2%$0.002$0.1724.1$0.37*–closed2026-09-22
Ling 3.0 Flash Sante (free)0.0%0.2%–––––closed2026-09-04
Inkling (free)0.0%0.2%–$1.7625.0–35Kopen2026-07-15
DeepSeek V4 Flash 07310.0%0.2%–$0.6634.3$0.22*–open2026-07-31
union-alpha0.0%0.2%–––––––
Ling 3.1 Flash0.0%0.1%$0.000$0.4541.1$1.00*–open2026-10-02
Ling 3.0 Flash VL0.0%0.1%–$0.1124.6––open2026-09-10

$ / task is one Artificial Analysis index task at OpenRouter's billed rate; * marks list price where no billed rate exists. Models with at least 0.1% of dollars or tokens.

How this is measured
  • Dollars are what OpenRouter billed. Weekly totals come from its spend, token and request charts. Per-model dollars come from its weekly leaderboards.
  • The charts name each week's top 9 models by spend and lump the rest into “Others”, shown here as Unattributed. Full per-model leaderboards start the week of Sep 14, so model and lab totals before then are floors.
  • Lab and open-weights shares use only weeks with at least 90% of dollars attributed: Sep 14 – Oct 11, 2026 (4 weeks).
  • Concentration, frontier lag and launch curves use each week's published top 9, the same basis every week.
  • App dollars and use-case dollars are estimates, marked est. or ≈. Provider shares come from OpenRouter's sampled provider mix, so they are shares, not billed dollars.
  • The range selector sets the weekly charts and the headline tiles. Snapshot sections name their own window, often the last 30 days. Weeks run Monday to Sunday; the current partial week is left out.
  • Scores, list prices and cost per task come from Artificial Analysis Intelligence Index v4.3.2. List price is its blend of 3 input tokens to 1 output token. Price of intelligence uses each month's last listed price.
Sources: OpenRouter rankings, leaderboards, model pages and model list; Artificial Analysis Intelligence Index. Updated Nov 22, 2026