BenchLeader
Menu

AI model usage: the most used models, labs and apps (Nov 2026)

Data as of 13 Nov 2026 · updated every six hours · by BenchLeader

DeepSeek V4.1 Flash 1017 is the most used model on OpenRouter this week, with 26.6% of tokens, and DeepSeek is the lab with the largest share (38.7%). This page lines up what people actually run, from OpenRouter, Vercel AI Gateway and coding-agent telemetry, against how the same models score on the BenchLeader Index.

Key statistics

Each statistic has its own link (#) to cite it directly.

  • As of 13 Nov 2026, DeepSeek V4.1 Flash 1017 is the most used model on OpenRouter by tokens over the past week (26.6% of all tokens), followed by Hy4 (9.7%) and DeepSeek V4.1 Flash (7.6%). #
  • DeepSeek models carried 38.7% of OpenRouter tokens in the week of 2 Nov 2026, the most of any lab, up 8.7 points on four weeks earlier. #
  • Open-weights models took 76.0% of OpenRouter tokens in the week of 2 Nov 2026, against 25.6% in the week of 17 Nov 2025. #
  • Of the ten most used models on OpenRouter this week, 2 are in the top 20 of the BenchLeader Index, and 0 have no Index score. #
  • On Vercel AI Gateway, Beam 1.0 had the largest share of tokens over the seven days to 12 Nov 2026 (25.5%). #
  • In coding agents run through T3 Code, Claude Code handled 67.1% of turns over the 30 days to 11 Nov 2026 and Codex 27.1%. #
  • Hermes Agent was the busiest app on OpenRouter over the past week, with 14.0 trillion tokens. #

Cite as: BenchLeader, “AI model usage”, data as of 13 Nov 2026, https://www.benchleader.com/usage.

Lab market share

Each lab's share of all tokens sent through OpenRouter, week by week for a year. OpenRouter names each week's nine busiest labs; everyone else is in one band, so a dash four weeks back in the table means the lab was not among that week's nine.

Share of OpenRouter tokens by lab, weeklyEach lab's share of all tokens processed through OpenRouter each week over the past year.Share of OpenRouter tokens by lab, weeklyDeepSeekGoogleAnthropicOpenAIZhipu AIAll other labs0%25%50%75%100%2026AprJulOctWeek of 17 Nov 2025: DeepSeek 5.3%, Google 20%, Anthropic 12%, OpenAI 9.9%, Zhipu AI 2.4%, All other labs 51%Week of 24 Nov 2025: DeepSeek 4.7%, Google 18%, Anthropic 11%, OpenAI 9.2%, Zhipu AI 2.1%, All other labs 55%Week of 1 Dec 2025: DeepSeek 6.5%, Google 21%, Anthropic 14%, OpenAI 11%, Zhipu AI 2.4%, All other labs 44%Week of 8 Dec 2025: DeepSeek 6.9%, Google 23%, Anthropic 15%, OpenAI 16%, Zhipu AI 2.3%, All other labs 36%Week of 15 Dec 2025: DeepSeek 8.1%, Google 23%, Anthropic 14%, OpenAI 15%, Zhipu AI 2.3%, All other labs 37%Week of 22 Dec 2025: DeepSeek 10%, Google 22%, Anthropic 13%, OpenAI 10%, Zhipu AI 2.5%, All other labs 41%Week of 29 Dec 2025: DeepSeek 11%, Google 23%, Anthropic 13%, OpenAI 9.8%, Zhipu AI 3.0%, All other labs 40%Week of 5 Jan 2026: DeepSeek 9.2%, Google 24%, Anthropic 17%, OpenAI 11%, Zhipu AI 3.4%, All other labs 35%Week of 12 Jan 2026: DeepSeek 8.3%, Google 26%, Anthropic 19%, OpenAI 13%, Zhipu AI 2.3%, All other labs 32%Week of 19 Jan 2026: DeepSeek 8.9%, Google 24%, Anthropic 17%, OpenAI 13%, Zhipu AI 2.5%, All other labs 35%Week of 26 Jan 2026: DeepSeek 9.6%, Google 24%, Anthropic 17%, OpenAI 14%, Zhipu AI 3.2%, All other labs 32%Week of 2 Feb 2026: DeepSeek 9.3%, Google 22%, Anthropic 15%, OpenAI 12%, Zhipu AI 0.0%, All other labs 42%Week of 9 Feb 2026: DeepSeek 7.6%, Google 17%, Anthropic 13%, OpenAI 11%, Zhipu AI 7.0%, All other labs 45%Week of 16 Feb 2026: DeepSeek 7.0%, Google 17%, Anthropic 14%, OpenAI 10%, Zhipu AI 7.6%, All other labs 45%Week of 23 Feb 2026: DeepSeek 7.5%, Google 20%, Anthropic 16%, OpenAI 11%, Zhipu AI 5.6%, All other labs 41%Week of 2 Mar 2026: DeepSeek 7.1%, Google 19%, Anthropic 16%, OpenAI 14%, Zhipu AI 0.0%, All other labs 44%Week of 9 Mar 2026: DeepSeek 7.3%, Google 18%, Anthropic 15%, OpenAI 12%, Zhipu AI 0.0%, All other labs 48%Week of 16 Mar 2026: DeepSeek 6.6%, Google 15%, Anthropic 14%, OpenAI 9.7%, Zhipu AI 7.5%, All other labs 47%Week of 23 Mar 2026: DeepSeek 6.5%, Google 13%, Anthropic 13%, OpenAI 9.6%, Zhipu AI 6.8%, All other labs 52%Week of 30 Mar 2026: DeepSeek 5.3%, Google 11%, Anthropic 10%, OpenAI 7.5%, Zhipu AI 4.7%, All other labs 62%Week of 6 Apr 2026: DeepSeek 7.1%, Google 16%, Anthropic 14%, OpenAI 12%, Zhipu AI 6.5%, All other labs 43%Week of 13 Apr 2026: DeepSeek 7.3%, Google 18%, Anthropic 18%, OpenAI 12%, Zhipu AI 7.0%, All other labs 38%Week of 20 Apr 2026: DeepSeek 7.7%, Google 16%, Anthropic 17%, OpenAI 12%, Zhipu AI 5.2%, All other labs 42%Week of 27 Apr 2026: DeepSeek 9.3%, Google 15%, Anthropic 14%, OpenAI 11%, Zhipu AI 4.2%, All other labs 47%Week of 4 May 2026: DeepSeek 12%, Google 14%, Anthropic 15%, OpenAI 10%, Zhipu AI 3.8%, All other labs 46%Week of 11 May 2026: DeepSeek 16%, Google 15%, Anthropic 15%, OpenAI 9.9%, Zhipu AI 3.5%, All other labs 41%Week of 18 May 2026: DeepSeek 20%, Google 15%, Anthropic 17%, OpenAI 10%, Zhipu AI 3.2%, All other labs 35%Week of 25 May 2026: DeepSeek 18%, Google 13%, Anthropic 18%, OpenAI 9.2%, Zhipu AI 3.0%, All other labs 39%Week of 1 Jun 2026: DeepSeek 19%, Google 12%, Anthropic 15%, OpenAI 8.0%, Zhipu AI 0.0%, All other labs 47%Week of 8 Jun 2026: DeepSeek 17%, Google 9.8%, Anthropic 15%, OpenAI 7.1%, Zhipu AI 0.0%, All other labs 50%Week of 15 Jun 2026: DeepSeek 19%, Google 8.9%, Anthropic 14%, OpenAI 8.1%, Zhipu AI 5.6%, All other labs 45%Week of 22 Jun 2026: DeepSeek 16%, Google 8.7%, Anthropic 14%, OpenAI 7.8%, Zhipu AI 5.7%, All other labs 48%Week of 29 Jun 2026: DeepSeek 18%, Google 9.3%, Anthropic 15%, OpenAI 7.9%, Zhipu AI 6.7%, All other labs 43%Week of 6 Jul 2026: DeepSeek 16%, Google 7.9%, Anthropic 14%, OpenAI 7.0%, Zhipu AI 7.3%, All other labs 48%Week of 13 Jul 2026: DeepSeek 14%, Google 6.8%, Anthropic 12%, OpenAI 6.8%, Zhipu AI 6.7%, All other labs 54%Week of 20 Jul 2026: DeepSeek 17%, Google 8.4%, Anthropic 9.1%, OpenAI 6.9%, Zhipu AI 6.5%, All other labs 52%Week of 27 Jul 2026: DeepSeek 22%, Google 8.5%, Anthropic 8.0%, OpenAI 9.8%, Zhipu AI 5.7%, All other labs 46%Week of 3 Aug 2026: DeepSeek 26%, Google 9.3%, Anthropic 7.0%, OpenAI 12%, Zhipu AI 5.5%, All other labs 40%Week of 10 Aug 2026: DeepSeek 26%, Google 9.0%, Anthropic 8.5%, OpenAI 13%, Zhipu AI 6.1%, All other labs 37%Week of 17 Aug 2026: DeepSeek 22%, Google 7.3%, Anthropic 5.2%, OpenAI 11%, Zhipu AI 4.4%, All other labs 50%Week of 24 Aug 2026: DeepSeek 19%, Google 7.5%, Anthropic 4.2%, OpenAI 11%, Zhipu AI 9.5%, All other labs 49%Week of 31 Aug 2026: DeepSeek 18%, Google 6.3%, Anthropic 4.9%, OpenAI 16%, Zhipu AI 16%, All other labs 39%Week of 7 Sep 2026: DeepSeek 19%, Google 5.9%, Anthropic 3.8%, OpenAI 19%, Zhipu AI 13%, All other labs 39%Week of 14 Sep 2026: DeepSeek 26%, Google 5.4%, Anthropic 3.6%, OpenAI 13%, Zhipu AI 15%, All other labs 38%Week of 21 Sep 2026: DeepSeek 23%, Google 4.7%, Anthropic 3.6%, OpenAI 13%, Zhipu AI 14%, All other labs 42%Week of 28 Sep 2026: DeepSeek 22%, Google 4.1%, Anthropic 3.9%, OpenAI 11%, Zhipu AI 8.7%, All other labs 50%Week of 5 Oct 2026: DeepSeek 30%, Google 3.9%, Anthropic 6.8%, OpenAI 9.9%, Zhipu AI 11%, All other labs 39%Week of 12 Oct 2026: DeepSeek 28%, Google 4.3%, Anthropic 8.1%, OpenAI 8.8%, Zhipu AI 11%, All other labs 39%Week of 19 Oct 2026: DeepSeek 32%, Google 3.4%, Anthropic 8.6%, OpenAI 9.1%, Zhipu AI 10%, All other labs 36%Week of 26 Oct 2026: DeepSeek 37%, Google 3.4%, Anthropic 7.4%, OpenAI 12%, Zhipu AI 11%, All other labs 30%Week of 2 Nov 2026: DeepSeek 39%, Google 3.0%, Anthropic 6.5%, OpenAI 8.9%, Zhipu AI 8.6%, All other labs 34%Week of 9 Nov 2026: DeepSeek 38%, Google 3.3%, Anthropic 7.3%, OpenAI 8.8%, Zhipu AI 8.4%, All other labs 34%last week still runningData: OpenRouter rankings (openrouter.ai/rankings),weekly tokens · data as of 13 Nov 2026www.benchleader.com/usage
Weekly share of OpenRouter tokens by lab, 17 Nov 2025 to 13 Nov 2026. Hover or tap a week for every lab's share; click a lab in the key to show or hide it.

Source: Usage: OpenRouter rankings · data as of 13 Nov 2026.

LabShare, week of 2 Nov 2026Four weeks earlierChangeVercel AI Gateway, last 7 days
DeepSeek38.7%30.0%+8.7 pts23.1%
Tencent9.0%6.3%+2.7 pts–
OpenAI8.9%9.9%−1.0 pts11.1%
Zhipu AI8.6%10.5%−1.9 pts14.0%
Anthropic6.5%6.8%−0.3 pts13.0%
NVIDIA3.9%5.0%−1.1 pts1.9%
StepFun3.5%5.2%−1.7 pts–
Moonshot AI3.1%––3.5%
Google3.0%3.9%−0.9 pts2.4%
All other labs14.8%14.4%+0.4 pts–

Source: Usage: OpenRouter rankings · Vercel AI Gateway, CC BY 4.0 · data as of 13 Nov 2026.

Cross-check on Vercel AI Gateway, share of tokens over the seven days to 12 Nov 2026: Reflection AI 25.5%, DeepSeek 23.1%, Zhipu AI 14.0%, Anthropic 13.0%, OpenAI 11.1%, Moonshot AI 3.5%. Two gateways with different customers: where they agree, the pattern is not one platform's quirk.

Most used models

Share of OpenRouter tokens over the past seven days (209.7 trillion in all), with each model's last 14 days and its change on the week before. The free and batch listings of a model count toward it. The BenchLeader Index sits beside each one: usage never enters the Index.

#ModelShareTokensvs week beforeLast 14 daysInput per output tokenIndex

Source: Usage: OpenRouter rankings · Index: BenchLeader · data as of 13 Nov 2026.

Most used over 30 days
#ModelShareTokensIndex
1DeepSeek V4.1 Flash 101721.3%176.4T58.1
2DeepSeek V4.1 Flash9.8%81.5T58.6
3GLM 5.3 Flash7.0%57.7T56.5
4GPT-6 Luna4.2%34.6T56.4
5Claude Opus 5.53.7%30.5T68.4
6MiniMax M3.1 Flash3.3%27.5T53.5
7Hy4 preview3.3%27.1T–
8Step 5 Preview3.1%26.0T54.3
9Hy42.9%24.3T56.7
10MiMo-V2.6-Flash2.5%20.6T54.8
11GLM 5.32.3%19.2T58.8
12DeepSeek V4 Flash2.0%16.8T57.1
13Nemotron 3 Ultra 550B A55B2.0%16.6T55.4
14GPT-6.1 Sol1.7%14.4T66.6
15Jev 1.131.6%12.9T–
16Muse Spark 1.31.5%12.2T61.5
17Hy31.4%12.0T55.7
18Claude Haiku 5.51.4%11.4T56.7
19DeepSeek V4 Flash 04231.3%11.1T–
20GPT-5.6 Luna1.3%11.1T58.2
21Nemotron 3.5 Super VL1.3%10.9T45.8
22Claude Sonnet 5.51.3%10.7T64.7
23Kimi K3.1 Pro1.1%9.49T61.3
24Kimi K30.91%7.58T61.6
25Jev 1.140.85%7.06T–

Source: Usage: OpenRouter rankings · Index: BenchLeader · data as of 13 Nov 2026.

How each model is used

What the requests to the most used models are made of over the last 30 days, from each model's own usage chart on OpenRouter: how much input per token of output, how much of the input was a cache hit, how much of the output was reasoning, and how many tool calls a request made. Availability is the share of requests served over the last 72 hours; tool-call errors are the share of tool calls that failed over the last seven days.

ModelTokens per requestInput per outputCached inputReasoning share of outputTool calls per requestAvailabilityTool-call errors
DeepSeek V4.1 Flash 101762.2k7294%66%0.5199.9%0.45%
Hy4141.5k9795%63%0.9399.8%0.14%
DeepSeek V4.1 Flash35.1k4690%67%0.2999.8%0.69%
GLM 5.3 Flash23.1k3183%38%0.2699.9%0.94%
GPT-6 Luna15.5k3276%56%0.2299.9%1.13%
Claude Opus 5.5124.2k11190%50%0.6499.9%0.47%
Step 5 Preview242.6k21095%6%0.1551.0%1.87%
Kimi K3.1 Pro115.9k11894%60%0.7499.9%0.10%
GPT-6.1 Sol38.6k5889%37%0.4289.9%0.24%
GLM 5.363.1k5991%72%0.5299.9%0.71%
Nemotron 3.5 Super VL42.5k2580%82%0.4399.9%2.69%
MiMo-V2.6-Flash66.0k7790%59%0.5799.6%2.10%
MiniMax M3.1 Flash87.0k6893%69%0.5999.8%3.07%
Crouton Alpha24.2k73–18%0.260.0%–
DeepSeek V4 Flash11.7k2172%50%0.1199.9%1.24%
Nemotron 3 Ultra 550B A55B10.0k1671%61%0.1699.9%3.17%
Muse Spark 1.370.5k7690%64%0.5599.0%1.27%
Claude Haiku 5.514.2k1863%54%0.2199.8%3.98%
Jev 1.133.1k12–––100.0%–
Claude Sonnet 5.548.8k5881%44%0.3799.9%2.68%

Source: Usage: OpenRouter rankings · data as of 13 Nov 2026.

What the tokens are for

OpenRouter sorts its traffic into kinds of work. Over the last 30 days: agent 32.7%, general 31.5%, code 28.7%, data 7.1% of tokens. The biggest single tasks, and the model that took the most of each:

TaskKindShare of tokensMost used model for it
Workflow executionAgent21.9%DeepSeek V4.1 Flash 1017 22.2%
Classification taggingGeneral11.5%Jev 1.13 12.1%
General implCode9.0%DeepSeek V4.1 Flash 1017 24.1%
Multi step planningAgent7.8%DeepSeek V4.1 Flash 1017 24.6%
DebuggingCode5.5%DeepSeek V4.1 Flash 1017 23.2%
ExtractionData5.3%GLM 5.3 Flash 13.4%
File read writeCode4.8%DeepSeek V4.1 Flash 1017 24.5%
Qa knowledgeGeneral3.7%Step 5 Preview 41.0%
SummarizationGeneral3.3%Step 5 Preview 78.3%
Review securityCode3.1%DeepSeek V4.1 Flash 1017 31.3%

Source: Usage: OpenRouter rankings · data as of 13 Nov 2026.

Usage against quality

The most used models on OpenRouter this week, placed by their share of tokens and their BenchLeader Index. Popular is not the same as best: cheap, fast models carry most of the traffic, and the highest-scoring models are a small share of it.

Usage vs quality: OpenRouter share vs IndexEach dot is one of the most used models on OpenRouter this week: its share of tokens (log scale) against its BenchLeader Index.Usage vs quality: OpenRouter share vs IndexOpen weightsClosed0.1%1%10%100%4045505560657075BenchLeader Index v1.4Share of OpenRouter tokens, past week (log)DeepSeek V4.1 Flash 1017: 58.1, 26.61%Hy4: 56.7, 9.66%DeepSeek V4.1 Flash: 58.6, 7.61%GLM 5.3 Flash: 56.5, 5.71%GPT-6 Luna: 56.4, 3.57%Claude Opus 5.5: 68.4, 3.24%Step 5 Preview: 54.3, 3.21%Kimi K3.1 Pro: 61.3, 2.09%GPT-6.1 Sol: 66.6, 2.02%GLM 5.3: 58.8, 1.85%Nemotron 3.5 Super VL: 45.8, 1.74%MiMo-V2.6-Flash: 54.8, 1.73%MiniMax M3.1 Flash: 53.5, 1.72%DeepSeek V4 Flash: 57.1, 1.54%Nemotron 3 Ultra 550B A55B: 55.4, 1.54%Muse Spark 1.3: 61.5, 1.43%Claude Haiku 5.5: 56.7, 1.34%Claude Sonnet 5.5: 64.7, 1.16%Hy3: 55.7, 1%GPT-5.6 Luna: 58.2, 0.79%Gemini 3.8 Flash: 61.7, 0.77%GPT-6.1 Terra: 64.1, 0.59%Kimi K3: 61.6, 0.54%Nemotron 3.5 Lightning: 41.4, 0.53%MiMo-V2.6-Pro: 56.8, 0.52%Gemini 4 Argon: 65.3, 0.48%GLM 5.2: 59.0, 0.44%Qwen3.8 27B: 53.1, 0.41%DeepSeek V4 Pro: 56.0, 0.37%MiniMax M3: 56.1, 0.31%Gemini 3 Flash: 58.9, 0.3%Inkling 2 Small: 57.9, 0.24%Gemini 3.1 Flash Lite: 52.4, 0.24%Grok 4.7: 58.0, 0.23%Claude Sonnet 5: 59.0, 0.22%Claude Fable 5.2: 72.7, 0.22%Gemma 4 31B: 49.8, 0.22%gpt-oss-120b: 45.8, 0.2%Gemma 4 26B A4B: 56.1, 0.2%Solar Pro 4: 48.9, 0.2%Gemini 2.5 Flash-Lite: 48.3, 0.15%DeepSeek V3.2: 56.3, 0.14%GPT-6 Astra: 67.1, 0.14%Gemini 3.7 Flash: 62.1, 0.14%Gemini 2.5 Flash: 52.8, 0.13%GPT-5.6 Sol: 65.3, 0.13%Muse Ember: 52.1, 0.12%DeepSeek V4.1 Flash 1017Hy4DeepSeek V4.1 FlashGLM 5.3 FlashGPT-6 LunaUsage: OpenRouter rankings, past weekIndex: BenchLeader v1.4 · data as of 13 Nov 2026www.benchleader.com/usage
Share of OpenRouter tokens over the past week (log scale) against the BenchLeader Index v1.4, for this week's most used models that have an Index score. The dashed lines are the medians of the models shown.

Source: Usage: OpenRouter rankings · Index: BenchLeader · data as of 13 Nov 2026.

Open weights against closed models

How OpenRouter's tokens split between models whose weights anyone can download and closed models served only by their maker, week by week. Compare the gap in quality on the AI race page.

Open weights vs closed, OpenRouter tokensWeekly share of OpenRouter tokens going to open-weights models and to closed models.Open weights vs closed, OpenRouter tokensOpen weightsClosed0%25%50%75%100%2026AprJulOctWeek of 17 Nov 2025: Open weights 26%, Closed 74%Week of 24 Nov 2025: Open weights 22%, Closed 78%Week of 1 Dec 2025: Open weights 27%, Closed 73%Week of 8 Dec 2025: Open weights 34%, Closed 66%Week of 15 Dec 2025: Open weights 36%, Closed 64%Week of 22 Dec 2025: Open weights 41%, Closed 59%Week of 29 Dec 2025: Open weights 42%, Closed 58%Week of 5 Jan 2026: Open weights 37%, Closed 63%Week of 12 Jan 2026: Open weights 34%, Closed 66%Week of 19 Jan 2026: Open weights 37%, Closed 63%Week of 26 Jan 2026: Open weights 38%, Closed 62%Week of 2 Feb 2026: Open weights 46%, Closed 54%Week of 9 Feb 2026: Open weights 54%, Closed 46%Week of 16 Feb 2026: Open weights 56%, Closed 44%Week of 23 Feb 2026: Open weights 50%, Closed 50%Week of 2 Mar 2026: Open weights 50%, Closed 50%Week of 9 Mar 2026: Open weights 50%, Closed 50%Week of 16 Mar 2026: Open weights 43%, Closed 57%Week of 23 Mar 2026: Open weights 39%, Closed 61%Week of 30 Mar 2026: Open weights 31%, Closed 69%Week of 6 Apr 2026: Open weights 42%, Closed 58%Week of 13 Apr 2026: Open weights 41%, Closed 59%Week of 20 Apr 2026: Open weights 47%, Closed 53%Week of 27 Apr 2026: Open weights 56%, Closed 44%Week of 4 May 2026: Open weights 57%, Closed 43%Week of 11 May 2026: Open weights 57%, Closed 43%Week of 18 May 2026: Open weights 58%, Closed 42%Week of 25 May 2026: Open weights 59%, Closed 41%Week of 1 Jun 2026: Open weights 65%, Closed 35%Week of 8 Jun 2026: Open weights 67%, Closed 33%Week of 15 Jun 2026: Open weights 68%, Closed 32%Week of 22 Jun 2026: Open weights 68%, Closed 32%Week of 29 Jun 2026: Open weights 68%, Closed 32%Week of 6 Jul 2026: Open weights 72%, Closed 28%Week of 13 Jul 2026: Open weights 75%, Closed 25%Week of 20 Jul 2026: Open weights 75%, Closed 25%Week of 27 Jul 2026: Open weights 73%, Closed 27%Week of 3 Aug 2026: Open weights 71%, Closed 29%Week of 10 Aug 2026: Open weights 69%, Closed 31%Week of 17 Aug 2026: Open weights 72%, Closed 28%Week of 24 Aug 2026: Open weights 72%, Closed 28%Week of 31 Aug 2026: Open weights 70%, Closed 30%Week of 7 Sep 2026: Open weights 67%, Closed 33%Week of 14 Sep 2026: Open weights 73%, Closed 27%Week of 21 Sep 2026: Open weights 70%, Closed 30%Week of 28 Sep 2026: Open weights 69%, Closed 31%Week of 5 Oct 2026: Open weights 73%, Closed 27%Week of 12 Oct 2026: Open weights 73%, Closed 27%Week of 19 Oct 2026: Open weights 74%, Closed 26%Week of 26 Oct 2026: Open weights 72%, Closed 28%Week of 2 Nov 2026: Open weights 76%, Closed 24%Week of 9 Nov 2026: Open weights 75%, Closed 25%last week still runningData: OpenRouter rankings (openrouter.ai/rankings)data as of 13 Nov 2026 · www.benchleader.com/usage
Weekly share of OpenRouter tokens to open-weights and closed models, 17 Nov 2025 to 13 Nov 2026.

Source: Usage: OpenRouter rankings · data as of 13 Nov 2026.

Top apps

The apps sending the most tokens, as each gateway ranks the apps that identify themselves to it. Coding agents dominate both lists.

OpenRouter, past week

#AppKindTokens
1Hermes AgentPersonal agent, CLI agent14.0T
2Claude CodeCLI agent11.8T
3Kilo CodeCLI agent, IDE extension6.56T
4piCLI agent4.45T
5CodexCLI agent4.02T
6OpenClawPersonal agent, CLI agent3.91T
7ompCLI agent3.15T
8DeepSeek Harness–2.58T
9StrixCLI agent2.55T
10Minds by Authentic Intelligence–1.63T
11Portkey AIProgramming app1.10T
12OpenHandsCLI agent1.04T
13Freebuff–792B
14ISEKAI ZEROGame745B
15Fleet Dashboard Post Helper–709B

Source: Usage: OpenRouter rankings · data as of 13 Nov 2026.

Vercel AI Gateway, by token volume

  1. 1Cline · Open source AI coding
  2. 2Kilo Code · AI coding agent for VS Code
  3. 3Command Code · AI coding agent that learns your taste
  4. 4opencode · AI coding agent built for the terminal
  5. 5Hermes Agent · Self-improving AI agent
  6. 6Zo Computer · Powerful AI in your personal cloud
  7. 7Goblins · Education AI app for tutoring
  8. 8Poke · AI assistant that lives in Apple Messages
  9. 9Ditto · Personalized AI matchmaking service
  10. 10Terac · AI interview platform

Source: Vercel AI Gateway, CC BY 4.0 · data as of 13 Nov 2026.

Coding agents: Claude Code, Codex and the rest

Which agents people drive, and which models inside them, from the telemetry of T3 Code, a desktop app that runs Claude Code, Codex, OpenCode and others. One app's users, not the whole market, and each day is weighted the same.

Claude Code vs Codex: gap in share of coding-agent turnsClaude Code's share of each day's coding-agent turns minus Codex's share, in percentage points.Claude Code vs Codex: gap in share of coding-agent turnsClaude Code aheadCodex ahead-60-40-200+20+40+6013 Oct23 Oct2 Nov11 NovClaude Code aheadCodex ahead13 Oct 2026: Claude Code 69.1%, Codex 25.5% (gap +43.6 points)14 Oct 2026: Claude Code 69.8%, Codex 24.8% (gap +45 points)15 Oct 2026: Claude Code 67.8%, Codex 26.7% (gap +41.1 points)16 Oct 2026: Claude Code 66.7%, Codex 27.2% (gap +39.5 points)17 Oct 2026: Claude Code 64.0%, Codex 29.5% (gap +34.6 points)18 Oct 2026: Claude Code 62.9%, Codex 30.9% (gap +32.1 points)19 Oct 2026: Claude Code 66.2%, Codex 27.8% (gap +38.4 points)20 Oct 2026: Claude Code 66.6%, Codex 27.7% (gap +38.8 points)21 Oct 2026: Claude Code 67.1%, Codex 27.6% (gap +39.5 points)22 Oct 2026: Claude Code 66.5%, Codex 28.3% (gap +38.3 points)23 Oct 2026: Claude Code 63.6%, Codex 31.0% (gap +32.6 points)24 Oct 2026: Claude Code 62.1%, Codex 32.1% (gap +30 points)25 Oct 2026: Claude Code 61.9%, Codex 32.2% (gap +29.7 points)26 Oct 2026: Claude Code 65.5%, Codex 28.7% (gap +36.8 points)27 Oct 2026: Claude Code 66.5%, Codex 27.7% (gap +38.7 points)28 Oct 2026: Claude Code 65.8%, Codex 28.7% (gap +37.1 points)29 Oct 2026: Claude Code 65.3%, Codex 29.1% (gap +36.2 points)30 Oct 2026: Claude Code 66.7%, Codex 27.5% (gap +39.2 points)31 Oct 2026: Claude Code 65.9%, Codex 27.8% (gap +38.1 points)1 Nov 2026: Claude Code 65.2%, Codex 28.5% (gap +36.8 points)2 Nov 2026: Claude Code 68.1%, Codex 25.9% (gap +42.1 points)3 Nov 2026: Claude Code 68.5%, Codex 25.5% (gap +43 points)4 Nov 2026: Claude Code 68.4%, Codex 25.6% (gap +42.8 points)5 Nov 2026: Claude Code 68.4%, Codex 25.6% (gap +42.8 points)6 Nov 2026: Claude Code 67.9%, Codex 26.4% (gap +41.5 points)7 Nov 2026: Claude Code 68.3%, Codex 25.4% (gap +42.9 points)8 Nov 2026: Claude Code 69.3%, Codex 24.2% (gap +45.1 points)9 Nov 2026: Claude Code 71.6%, Codex 22.1% (gap +49.5 points)10 Nov 2026: Claude Code 72.7%, Codex 21.2% (gap +51.5 points)11 Nov 2026: Claude Code 73.9%, Codex 20.1% (gap +53.8 points)Data: T3 Code telemetry via Slopalytics(slopalytics.com/usage) · data as of 12 Nov 2026www.benchleader.com/usage
Claude Code's share of each day's turns minus Codex's, in percentage points, 13 Oct 2026 to 11 Nov 2026. Hover a day for both shares.

Source: T3 Code via Slopalytics · data as of 12 Nov 2026.

Share of turns by agent

Agent30 daysFirst weekLast week
Claude Code67.1%66.7%70.3%
Codex27.1%27.5%23.6%
OpenCode2.3%2.5%2.1%
Pi1.0%1.0%1.1%
Other0.90%0.80%1.1%
Cursor0.70%0.60%0.70%
Antigravity0.60%0.50%0.80%

Models inside coding agents

ModelShare of turnsMostly in
Claude Opus 5.557.1%Claude Code (99%)
GPT-6.1 Sol19.0%Codex (99%)
GPT-6 Astra5.0%Codex (99%)
Claude Sonnet 5.53.8%Claude Code (98%)
Claude Fable 5.11.9%Claude Code (100%)
GPT-6 Luna1.2%Codex (97%)
Claude Haiku 5.50.86%Claude Code (98%)
GPT-6.1 Terra0.74%Codex (99%)
Claude Fable 5.20.62%Claude Code (99%)
GPT-6 Sol0.20%Codex (98%)
GPT-5.6 Sol0.17%Codex (93%)
Claude Opus 50.14%Claude Code (99%)

Source: T3 Code via Slopalytics · data as of 12 Nov 2026.

What a turn looks like

A turn is one prompt run to its end, every tool call and re-read of the context included. Cost is Slopalytics' estimate at API list prices for the models it can price, not what anyone was billed.

Input per turn
2,940k
98% from cache
Output per turn
10,035
tokens
Estimated cost per turn
$1.18
API list prices
Time per completed turn
314 s
97% completed
Per turnAll agentsClaude CodeCodexOpenCode
Input tokens2,940k2,821k3,112k4,242k
Output tokens10,0359,18411,49117,534
Input from cache97.8%98.2%97.4%95.7%
Estimated cost$1.181$1.027$2.548$1.563
Subagents per 100 turns11.311.611.73.8
Completed96.5%97.3%96.2%88.5%
Interrupted or cancelled1.5%1.0%1.6%4.8%
Failed2.0%1.7%2.2%6.7%
Seconds per completed turn314256428520

Source: T3 Code via Slopalytics · data as of 12 Nov 2026.

Which plan is the cheapest way to run these agents: AI coding subscriptions compared.

Which provider is fastest for popular models

For the most used models, the fastest host each gateway measured. OpenRouter's figures are medians across its providers over recent traffic; Vercel's are each host's median over the last hour and uptime over the last day; Hugging Face's are the router's latest reading for open-weights models.

ModelOpenRouter medianOpenRouter fastestVercel fastestVercel best uptimeHugging Face fastest
DeepSeek V4.1 Flash 1017446 tok/s · 0.26 sBaseTen (of 48)DigitalOcean · 471 tok/sDigitalOcean · 100.00%Novita · 114 tok/s · 0.95 s
Hy447 tok/s · 3.50 sTencentTencent · 95 tok/sTencent · 100.00%DeepInfra · 56 tok/s · 0.59 s
DeepSeek V4.1 Flash271 tok/s · 0.20 sVenice (of 37)Parasail · 381 tok/sParasail · 100.00%Baseten · 182 tok/s · 0.22 s
GLM 5.3 Flash144 tok/s · 0.29 sOpenInference (of 42)Runware · 309 tok/sFriendli · 100.00%Baseten · 82 tok/s · 0.40 s
GPT-6 Luna131 tok/s · 0.88 sAmazon Bedrock (of 8)Amazon Bedrock · 189 tok/sOpenAI · 99.98%–
Claude Opus 5.5182 tok/s · 0.99 sAnthropic (of 11)Anthropic · 293 tok/sAmazon Bedrock · 99.99%–
Step 5 Preview104 tok/s · 3.49 sStepFun (of 2)Tencent · 90 tok/sTencent · 100.00%–
Kimi K3.1 Pro110 tok/s · 0.59 sDigitalOcean (of 24)DigitalOcean · 208 tok/sNebius · 100.00%DeepInfra · 36 tok/s · 2.59 s
GPT-6.1 Sol79 tok/s · 1.89 sOpenAI (of 9)Amazon Bedrock · 119 tok/sAzure · 99.93%–
GLM 5.3151 tok/s · 0.31 sBaseTen (of 46)DigitalOcean · 229 tok/sInceptron · 100.00%Nscale · 109 tok/s · 0.53 s
Nemotron 3.5 Super VL337 tok/s · 0.45 sVenice (of 11)Baseten · 710 tok/sDeepInfra · 100.00%–
MiMo-V2.6-Flash77 tok/s · 0.58 sCohere (of 8)DeepInfra · 65 tok/sTencent · 99.94%Novita · 21 tok/s · 1.48 s
MiniMax M3.1 Flash274 tok/s · 0.38 sInferact vLLM (of 12)Novita · 316 tok/sNovita · 100.00%Novita · 103 tok/s · 1.73 s
DeepSeek V4 Flash100 tok/s · 0.25 sParasail (of 13)Azure · 241 tok/sAlibaba Cloud · 100.00%DeepInfra · 83 tok/s · 0.21 s
Nemotron 3 Ultra 550B A55B125 tok/s · 0.22 sKT (of 6)DeepInfra · 172 tok/sBaseten · 99.97%DeepInfra · 85 tok/s · 0.48 s
Muse Spark 1.370 tok/s · 3.59 sMetaMeta · 89 tok/sMeta · 97.37%–
Claude Haiku 5.5180 tok/s · 0.43 sGoogle (of 10)Claude on AWS · 296 tok/sGoogle Vertex · 100.00%–
Claude Sonnet 5.5117 tok/s · 0.84 sAzure (of 10)Claude on AWS · 154 tok/sGoogle Vertex · 99.99%–
Hy347 tok/s · 1.57 sNovita (of 4)Novita · 161 tok/sNovita · 100.00%DeepInfra · 42 tok/s · 0.64 s
GPT-5.6 Luna143 tok/s · 0.47 sAmazon Bedrock (of 8)Amazon Bedrock · 363 tok/sAzure · 99.99%–

Source: Usage: OpenRouter rankings · Vercel AI Gateway, CC BY 4.0 · Hugging Face Inference Providers · data as of 13 Nov 2026.

Speed and price for every model across providers: fastest LLMs and LLM speed statistics.

How these numbers are made

OpenRouter figures are read from the public endpoints behind openrouter.ai/rankings: tokens are input plus output, a model's free and batch listings count toward it, and shares are of everything OpenRouter served, including models BenchLeader does not track. The lab chart folds every lab outside a week's nine busiest into one band, so a smaller lab shows as zero in weeks it missed the cut.

Vercel AI Gateway publishes its leaderboard as daily shares under CC BY 4.0; the shares here are seven-day averages of its daily figures. T3 Code's telemetry, as Slopalytics publishes it, scales every day to the same number of turns, so shares are real but turn counts are not volumes. Hugging Face's router lists each host's latest first-token latency and throughput.

Every gateway sees its own customers, not the market. None of these numbers enters the BenchLeader Index: a popular model is not thereby a good one. Names link to a model page only when BenchLeader tracks that model; other names are shown as the gateway gives them.

Questions people ask

What is the most used AI model right now?
On the model marketplace OpenRouter, DeepSeek V4.1 Flash 1017 carried 26.6% of all tokens over the seven days to 13 Nov 2026, the most of any model. Hy4 was second with 9.7%. On Vercel AI Gateway the leader by tokens was Beam 1.0 (25.5% over the last seven days).
Which AI lab's models are used the most?
DeepSeek, with 38.7% of OpenRouter tokens in the week of 2 Nov 2026, ahead of Tencent (9.0%) and OpenAI (8.9%).
Is Claude Code or Codex used more?
In T3 Code's telemetry, Claude Code ran 67.1% of coding-agent turns over the 30 days to 11 Nov 2026 and Codex 27.1%. That is one desktop app's users, not the whole market.
Does usage count toward the BenchLeader Index?
No. The Index is built only from independent benchmark results. Usage is shown beside it, because the most used models are mostly cheap and fast ones rather than the highest scoring.
Where does this usage data come from?
OpenRouter's public rankings, Vercel AI Gateway's leaderboard export (CC BY 4.0), T3 Code's telemetry as published by Slopalytics, and Hugging Face's inference router. Each gateway sees only its own customers, so each table names its source.

Related: most used models in coding agents, the leaderboard today, AI trends and statistics, the AI race, AI coding subscriptions, AI providers.