Find the right model with data.
Explore models through real usage and independent benchmarks. Select a ranking, search, sort and compare.
Developer adoption · OpenRouter
Based on real token usage routed through the OpenRouter API and collected daily. It measures adoption rather than quality and covers OpenRouter traffic, not the entire market.
Model usage Top10
DeepSeek+55%
DeepSeek+55%
Zhipu AI+25%
Zhipu AI+25%DeepSeek-22%
DeepSeek-22%
OpenAI+36%
OpenAI+36%
Tencent+5%
Tencent+5%
Anthropic-16%
Anthropic-16%
Stepfun+123%
Stepfun+123%
MiniMax+1324%
MiniMax+1324%Zhipu AI-12%
Zhipu AI-12%OpenAI+146%
OpenAI+146%Methodology and update notes ＋
Ranked by OpenRouter API actual token processing volume (7 daily rolling window); it reflects developer adoption rather than quality, and does not reflect overall market traffic. The month-over-month figure is the change compared to the previous 7 daily window token volume.
Source: OpenRouter (openrouter.ai/rankings), as of 2026-11-01T04:39:39.884Z · CC BY 4.0
Provider share
+4.9pp
+4.9pp+0.8pp
+0.8pp+2.7pp
+2.7pp-1.4pp
-1.4pp0pp
0pp+2pp
+2pp
-1.3pp
-5.1pp
-5.1pp
-0.3pp
-0.3pp
0pp
0ppMethodology and update notes ＋
Ranked by token share aggregated by permaslug vendor model prefix (7 daily rolling window). The methodology is top-50 approximation: the denominator includes other long-tail aggregate rows, and models not appearing on the list are counted as part of the long tail.
Programming Agent session cost
Click a column to sort. Select up to 3 entries to compare within this ranking.
| Comparative selection | ||||
|---|---|---|---|---|
 Codex | Muse Spark 1.3 Contributor 20260902 | 10-49-turns | $0.01 | |
| Codex | GLM 5.3 Flash 20260826 | 10-49-turns | $0.03 | |
| Codex | GPT 6 Luna 20260922 | 10-49-turns | $0.03 | |
 Claude Code | DeepSeek V4 Flash 20260423 | 10-49-turns | $0.02 | |
| Claude Code | GPT 6 Luna 20260922 | 10-49-turns | $0.03 | |
| Claude Code | MiMo V2.6 Flash 20260921 | 10-49-turns | $0.03 | |
| DeepSeek V4.1 Flash 20261017 | 10-49-turns | $0.03 | ||
| Qwen3.8 Flash 20260826 | 10-49-turns | $0.03 | ||
| GPT 4o Mini | 10-49-turns | $0.03 |
No matching records.
Methodology and update notes ＋
Programming Agent cost per session, median (USD, paid traffic), aggregated by harness × model × rounds, with source data refreshing weekly. The primary metric is the typical working session across 10-49 rounds, and each group of harness shows the lowest cost among the 3 model combinations for that tier.
Top Apps · App Rankings
Ranked by token consumption volume
Click a column to sort. Select up to 3 entries to compare within this ranking.
| Comparative selection | ||||
|---|---|---|---|---|
| 1 | 12.91T | 120441.1k | ||
| 2 | Claude Code | 10.04T | 100023.1k | |
| 3 |  pi | 5.48T | 43419.4k | |
| 4 | Codex | 5.23T | 75724.1k | |
| 5 |  Kilo Code | 4.81T | 45759.2k | |
| 6 |  OpenClaw | 3.58T | 26952.3k | |
| 7 |  omp | 1.93T | 12552.9k | |
| 8 |  DeepSeek Harness | 1.84T | 12473.3k | |
| 9 |  Strix | 1.76T | 15019.5k | |
| 10 |  Freebuff | 1.25T | 10356.4k |
No matching records.
Ranked by growth trend
Click a column to sort. Select up to 3 entries to compare within this ranking.
| Comparative selection | ||||
|---|---|---|---|---|
| 1 | Codex | 5.23T | 75724.1k | |
| 2 | pi | 5.48T | 43419.4k | |
| 3 | OpenClaw | 3.58T | 26952.3k | |
| 4 | 12.91T | 120441.1k | ||
| 5 | Strix | 1.76T | 15019.5k | |
| 6 |  Portkey AI | 1.14T | 102359.2k | |
| 7 | Aura | 0.43T | 3869.1k | |
| 8 | Nous Research API | 0.42T | 8536.1k | |
| 9 | LangChain | 0.51T | 12146.5k | |
| 10 |  CodeGPT | 0.43T | 2490k |
No matching records.
Methodology and update notes ＋
The public application token consumption ranking routed via OpenRouter (7 daily window). popular is ranked by consumption volume; trending is ranked by absolute excess growth (window volume minus the average volume of the preceding three equal-length periods), and applications with zero growth are not included.
Capability benchmarks
Five independent benchmarks provide complementary views: user preference (LMArena), general intelligence (AA), Chinese-language tasks (SuperCLUE), software engineering (SWE-bench) and agent tasks (Terminal-Bench).
Arena Text · Overall
Click a column to sort. Select up to 3 entries to compare within this ranking.
| Comparative selection | |||||
|---|---|---|---|---|---|
| 1 | Gemini 4 Carbon (High) | google | 1540 | ||
| 2 | Gemini 4 Argon (High) | google | 1528.6 | ||
| 3 | Claude Opus 5.5 (High) | anthropic | 1515.8 | ||
| 4 | Claude Fable 5.1 (Max) | anthropic | 1513.8 | ||
| 5 | Claude Opus 5 (Max) | anthropic | 1507.9 | ||
| 6 | Claude Opus 5 (High) | anthropic | 1504.1 | ||
| 7 | Claude Opus 4.6 (High) | anthropic | 1501.7 | ||
| 8 | Muse Wonder (Max) |  meta | 1499.2 | ||
| 9 | Gemini 3.8 Flash (High) | google | 1498.5 | ||
| 10 | Claude Opus 4.6 | anthropic | 1496.2 |
No matching records.
Methodology and update notes ＋
Official Text Arena overall leaderboard. Showing the first 10 of 419 records. The date is the upstream publication date; scores are not comparable across benchmarks.
Source: Arena · CC BY 4.0 · Collected 2026-11-01T04:15:16.955914Z
AA Intelligence Index
Click a column to sort. Select up to 3 entries to compare within this ranking.
| Comparative selection | |||||
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (max with fallback) | Anthropic | 58 | ||
| 2 | Claude Sonnet 5.5 (max with fallback) | Anthropic | 56 | ||
| 3 | Claude Opus 5.5 (xhigh with fallback) | Anthropic | 56 | ||
| 4 | Muse Wonder Full (max) | Meta | 56 | ||
| 5 | Kimi K3.1 Pro | 55 | |||
| 6 | Claude Opus 5.5 (high with fallback) | Anthropic | 54 | ||
| 7 | Claude Fable 5.1 (max with fallback) | Anthropic | 53 | ||
| 8 | Claude Fable 5.1 (xhigh with fallback) | Anthropic | 53 | ||
| 9 | GPT-6 Astra (max) | OpenAI | 53 | ||
| 10 | Gemini 4 Argon (high) | Google | 53 |
No matching records.
Methodology and update notes ＋
2026-10-26 Re-check the current top 10 model/inference settings of the official Artificial Analysis Intelligence Index; this observation date is not the model evaluation release date.
This ranking is manually maintained, with data as of 2026-10-26.
SuperCLUE Intelligent Index (Chinese)
Click a column to sort. Select up to 3 entries to compare within this ranking.
| Comparative selection | |||||
|---|---|---|---|---|---|
| 1 | Qwen3.8-Max-0902(max) | 72.62 | |||
| 1 | DeepSeek-V4.1-Flash(max) | DeepSeek | 71.81 | 9 monthly addition | |
| 2 | Qwen3.8-Max(max) | 71.48 | |||
| 2 | GLM-5.3(max) | Zhipu AI | 71.29 | ||
| 2 | Kimi-K3(max) | 70.68 | |||
| 3 | DeepSeek-V4-Pro-0813(max) | DeepSeek | 70.1 | ||
| 4 | GLM-5.3-Flash(max) | Zhipu AI | 68.1 | ||
| 5 | Qwen3.8-Flash(max) | 66.41 | |||
| 5 | Doubao-Seed-2.1-pro(high) |  ByteDance | 65.94 | ||
| 5 | DeepSeek-V4-Flash(max) | DeepSeek | 65.6 |
No matching records.
Methodology and update notes ＋
Six new-task monthly evaluations for the Chinese scene tasks (mathematical/scientific reasoning, code generation, precise instruction following, hallucination control, agent task planning) are conducted, with a maximum score of 100. The data is drawn from the official website's overall leaderboard for 2026 year 7 month; this period's overall leaderboard only includes domestic models, and overseas models do not participate in the ranking.
This ranking is manually maintained, with data as of 2026-09-28.
SWE-bench Verified
Click a column to sort. Select up to 3 entries to compare within this ranking.
| Comparative selection | |||||
|---|---|---|---|---|---|
| 1 | Claude 4.5 Opus (high) | Anthropic | 76.8 | mini-SWE-agent 2.0.0 | |
| 2 | Gemini 3 Flash (high) | Google | 75.8 | mini-SWE-agent 2.0.0 | |
| 3 | MiniMax M2.5 (high) | MiniMax | 75.8 | mini-SWE-agent 2.0.0 | |
| 4 | Claude 4.6 Opus | Anthropic | 75.6 | mini-SWE-agent 2.0.0 | |
| 5 | Claude 4.5 Opus (20251101) (medium) | Anthropic | 74.4 | mini-SWE-agent 1.16.0 | |
| 6 | Gemini 3 Pro Preview (2025-11-18) | Google | 74.2 | mini-SWE-agent 1.15.0 | |
| 7 | GLM 5 (high) | Zhipu AI | 72.8 | mini-SWE-agent 2.0.0 | |
| 8 | GPT 5.2 (high) | OpenAI | 72.8 | mini-SWE-agent 2.0.0 | |
| 9 | GPT 5.2 Codex | OpenAI | 72.8 | mini-SWE-agent 2.0.0 | |
| 10 | GPT 5.2 (2025-12-11) (high) | OpenAI | 71.8 | mini-SWE-agent 1.17.2 |
No matching records.
Methodology and update notes ＋
2026-10-26 Re-check the top 10 items of Verified and mini-SWE-agent (Bash Only) under the official JSON; retain the official model/inference setting names and the agent version of each row, and do not mix in other agent environments. This observation date is not the evaluation release date.
This ranking is manually maintained, with data as of 2026-10-26.
Terminal-Bench 2.1
Click a column to sort. Select up to 3 entries to compare within this ranking.
| Comparative selection | |||||
|---|---|---|---|---|---|
| 1 | GPT-6 Astra (high) | OpenAI | 87.4 | Codex | |
| 2 | GPT-6 Astra (medium) | OpenAI | 87 | Codex | |
| 3 | GPT-6 Astra (low) | OpenAI | 86.7 | Codex | |
| 3 | GPT-6 Astra (max) | OpenAI | 86.7 | Codex | |
| 5 | GPT-6 Astra (xhigh) | OpenAI | 85.8 | Codex | |
| 6 | Fable 5 (xhigh) | Anthropic | 83.8 | Claude Code | |
| 7 | GPT-5.5 (xhigh) | OpenAI | 83.2 | Codex | |
| 8 | Fable 5 (high) | Anthropic | 80.5 | Terminus 2 | |
| 9 | Grok 4.5 (high) |  xAI | 79.3 | Cursor CLI | |
| 10 | Opus 4.8 (high) | Anthropic | 78.9 | Claude Code |
No matching records.
Methodology and update notes ＋
2026-10-26 Re-check the top 10 model/agent/effort settings of the official Harbor Terminal-Bench 2.1 rev.6 leaderboard; retain the 2.1 baseline and do not mix in 4.0. This observation date is not the evaluation release date.
This ranking is manually maintained, with data as of 2026-10-26.