AI NEXTIA / MODEL INTELLIGENCE
Which AI models are people actually using?
A clearer view of the AI landscape—usage, adoption, and independent capability evidence. No invented composite scores. No ranking without a source.
Measured by tokens on a single routing platform, not global model adoption.
01 / ROUTED MODEL USAGE
Most used models on OpenRouter.
Actual token volume processed by model variants on the OpenRouter platform—one important window into developer/API demand.
Developer & API usage
Ranked by prompt + completion tokens, as recorded by OpenRouter. Individual versions and variants remain separate.
Original leaderboard ↗| Rank | Model / provider | Relative volume | Tokens | Change |
|---|---|---|---|---|
| #01 | DeepSeek V4.1 Flash 1017DeepSeek | 121T | new | |
| #02 | DeepSeek V4.1 FlashDeepSeek | 106T | +37% | |
| #03 | GLM 5.3 FlashZ.ai | 58.5T | +6% | |
| #04 | Hy4 previewTencent | 35T | -27% | |
| #05 | Step 5 PreviewStepFun | 34.3T | new | |
| #06 | GPT-6 LunaOpenAI | 33.9T | +211% | |
| #07 | Claude Opus 5.5Anthropic | 29.4T | +533% | |
| #08 | MiMo-V2.6-FlashXiaomi | 28.6T | +57% | |
| #09 | Nemotron 3 Ultra (free)NVIDIA | 20.7T | -2% | |
| #10 | GLM 5.3Z.ai | 19.5T | +59% |
| Rank | Model / provider | Relative volume | Tokens | Change |
|---|---|---|---|---|
| #01 | DeepSeek V4.1 Flash 1017DeepSeek | 6.9T | -32% | |
| #02 | Hy4Tencent | 2.72T | +110% | |
| #03 | DeepSeek V4.1 FlashDeepSeek | 2.57T | -14% | |
| #04 | GLM 5.3 FlashZ.ai | 1.64T | -15% | |
| #05 | Step 5 PreviewStepFun | 1.11T | +9% | |
| #06 | GPT-6 LunaOpenAI | 1.08T | -5% | |
| #07 | Claude Opus 5.5Anthropic | 1.04T | -10% | |
| #08 | Crouton AlphaStealth | 973B | +71% | |
| #09 | MiniMax M3.1 FlashMiniMax | 657B | -12% | |
| #10 | GPT-6.1 SolOpenAI | 641B | +8% |
High token volume means extensive use within OpenRouter. It does not mean most global users, highest quality, or lowest price. Different tokenizers and model verbosity also affect totals. The 30-day view is the better trend signal; the daily view is more volatile.
AI NEXTIA / AUTOMATED DISCOVERY
Recent model and platform announcements.
Getting started with Web Answer on Amazon Bedrock AgentCore
AWS Machine Learning Blog · Unreviewed headline
Original publication ↗Models / Nov 5How Amazon Marketplace evaluates content quality with Amazon SageMaker AI
AWS Machine Learning Blog · Unreviewed headline
Original publication ↗Models / Nov 5Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips
Google DeepMind · Unreviewed headline
Original publication ↗Models / Nov 5Australian universities are creating AI-native graduates
OpenAI News · Unreviewed headline
Original publication ↗Models / Nov 5Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips
Google The Keyword — AI · Unreviewed headline
Original publication ↗Automated external headlines, not independent editorial verification. Last checked: Nov 6, 05:17 UTC.
02 / CONSUMER PRODUCTS
Where everyday AI users go.
A different lens: popular AI websites, not specific model variants.
The most visited AI site is not necessarily the most-used model.
ChatGPT, Gemini and Claude are consumer products. A user can interact with different models inside an app, often without knowing which version is running.
Meanwhile, API-routing charts reflect developers and applications using models behind the scenes. Both perspectives matter; neither represents the complete AI market.
Explore Similarweb's ranking ↗Source: Similarweb, September 2026, updated October 1, 2026. This table uses website rank only; it does not infer user counts or subscription revenue. OpenAI.com is a company website in this category, not a direct measure of ChatGPT usage.
03 / INDEPENDENT EVALUATION
Popularity isn't performance.
Selected leading configurations from the Artificial Analysis Intelligence Index. These are capability benchmark scores, not usage totals.
Gemini 4 Carbon (High)
Claude Opus 5.5 (Max)
Anthropic
GPT-6 Astra (Max)
OpenAI
Source: Artificial Analysis, November 5, 2026 review. These are selected leading model families/configurations—not a complete leaderboard. Different inference settings and methodology changes affect scores. See the full source for the current rankings.
04 / MODEL ECOSYSTEM
Meet the organizations behind the models.
Browse major model makers, beyond the best-known American consumer assistants.
OpenAI
ChatGPT and GPT models
Official site ↗Assistant · APIAnthropic
Claude and coding agents
Official site ↗Assistant · APIGemini and DeepMind research
Official site ↗Models · APIDeepSeek
Open and hosted language models
Official site ↗Models · APIZ.ai
GLM language models
Official site ↗Models · APIQwen
Alibaba's Qwen model family
Official site ↗Assistant · APIMoonshot AI
Kimi models and assistants
Official site ↗Models · APIMiniMax
Multimodal and language models
Official site ↗Models · APITencent
Hunyuan model family
Official site ↗Research · ModelsXiaomi
MiMo model family
Official site ↗Models · InfrastructureNVIDIA
Nemotron models and inference
Official site ↗Research · ModelsMeta
Open-weight model research
Official site ↗Assistant · APIxAI
Grok models and products
Official site ↗Models · ResearchStepFun
Step model family
Official site ↗05 / METHODOLOGY
What each ranking really means.
Usage ≠ quality
OpenRouter shows public token processing on one API marketplace. Model variants are separated, not combined into global family totals.
Visits ≠ users
Similarweb provides website-traffic estimates. A person may visit repeatedly, and enterprise/API usage is largely absent.
Benchmarks ≠ your tasks
Artificial Analysis evaluates technical performance under a defined methodology. Your workload may have different accuracy, cost, and latency needs.
GET BETTER AT MODEL CHOICE
How to evaluate AI models for your actual work.
Start with representative tasks, not a single global ranking. Measure failure cases, price, latency, access requirements, and total cost of human review.
INTELLIGENCE FOR WHAT'S NEXT
A model landscape that keeps moving. A resource that keeps its sources visible.
Data refreshes automatically after the publisher configures an OpenRouter API key. Without one, this page honestly displays its dated editorial snapshot.