modelgrep

Most-used AI models, by task

Match · Updated 2026-08-04

DeepSeek V4 Flash 0423 is the most-used model on OpenRouter overall, leading 19 of 29 tracked tasks over the last 7 days.

29tasks tracked
7dwindow

Every other ranking here answers which model is best. This one answers which model people actually use — measured on real production traffic rather than a benchmark harness. The two disagree more often than you would expect, because adoption is decided by price, latency and availability at least as much as by capability.

Benchmark leaders aren't the models people use
We hold benchmarks and traffic together, so the gap is measurable — and it is nearly total.
Read the analysis →

Most-used models overall

1deepseek logoDeepSeek V4 Flash 042328.1%
2google logoGemini 2.5 Flash Lite10.3%
3openai logoGPT-5.6 Luna7.3%
4google logoGemini 2.5 Flash6.7%
5google logoGemini 3.1 Flash Lite4.4%
6openai logogpt-oss-120b4.3%
7google logoGemini 3 Flash Preview4.0%
8deepseek logoDeepSeek V4 Pro3.9%
9openai logoGPT-4o-mini3.9%
10ILing-2.6-flash3.2%

Share of classified requests, weighted by how much traffic each task represents.

code

Writing, debugging, reviewing and running code · 9.3% of requests
Code Generation
2.4%
#1 MiMo-V2.5 · 8.7%
Debugging
1.9%
#1 DeepSeek V4 Flash 0423 · 11.1%
File I/O
1.4%
#1 DeepSeek V4 Flash 0423 · 11.5%
Shell Execution
1.2%
#1 DeepSeek V4 Pro · 24.1%
Code Review
0.90%
#1 GLM 5.2 · 20.6%
Frontend & UI
0.70%
#1 MiMo-V2.5 · 16.4%
Repo Scanning
0.50%
#1 MiMo-V2.5 · 17.1%
SQL & Database
0.20%
#1 DeepSeek V4 Flash 0423 · 10.5%
DevOps & Config
0.10%
#1 MiMo-V2.5 · 12.5%

agent

Multi-step planning, tool use and workflow execution · 12.3% of requests
Workflow Execution
7.3%
#1 DeepSeek V4 Flash 0423 · 13.1%
Multi-step Planning
2.5%
#1 DeepSeek V4 Flash 0423 · 17.1%
Memory Extraction
1.1%
#1 DeepSeek V3.1 Terminus · 21.1%
Tool Dispatch
0.70%
#1 DeepSeek V4 Flash 0423 · 22.5%
Web Search
0.70%
#1 DeepSeek V4 Flash 0423 · 24.3%

data

Extracting and reshaping structured data · 18.4% of requests
Data Extraction
14.7%
#1 DeepSeek V4 Flash 0423 · 13.9%
Data Transformation
3.7%
#1 DeepSeek V4 Flash 0423 · 21.9%

general

Everything else — chat, writing, translation, classification · 60.0% of requests
Classification
26.8%
#1 DeepSeek V4 Flash 0423 · 11.9%
Roleplay & Fiction
12.4%
#1 DeepSeek V4 Flash 0423 · 42.2%
Content Writing
5.7%
#1 DeepSeek V4 Flash 0423 · 15.6%
Translation
2.9%
#1 DeepSeek V4 Flash 0423 · 20.7%
Conversation
2.8%
#1 DeepSeek V4 Flash 0423 · 18.2%
Q&A & Knowledge
2.7%
#1 DeepSeek V4 Flash 0423 · 13.1%
Summarization
2.7%
#1 gpt-oss-120b · 15.9%
Customer Support
2.1%
#1 Gemini 2.5 Flash · 22.6%
Research & Reports
0.60%
#1 DeepSeek V4 Flash 0423 · 12.4%
Finance & Trading
0.60%
#1 DeepSeek V4 Flash 0423 · 11.1%
Security Audit
0.30%
#1 DeepSeek V4 Flash 0423 · 11.7%
Math
0.30%
#1 GLM 5.2 · 8.9%
DevOps
0.10%
#1 DeepSeek V4 Flash 0423 · 12.2%

Source: OpenRouter (openrouter.ai/rankings), as of 2026-08-04. Shares are of classified, sampled traffic over a trailing 7-day window; absolute volumes are not published.