The most-used model in the AI coding tool Cline has changed. The numbers from Cline's official account are clear: DeepSeek V4-Flash is the most-used of all models. At the same time, one developer's account of canceling a $200-a-month Claude subscription passed 160,000 views and spread.
Four Records in a Row: The Swap in Numbers
The figures Cline released split at the July 31 update.
| Metric | Change after the 0731 update |
|---|---|
| Daily usage | +40% |
| Token usage | 3x |
| Rank | No. 1 of all models, with more users than the 2nd and 3rd combined |
| Record | All-time high four days in a row |
Cline is a tool where developers choose which model to attach to their coding agent. In it, one model being used more than the second and third combined means real-world use, not benchmark scores, has split the field.
The Person Who Cut $200 a Month: It Was the Combination, Not the Price
There is a case that shows the same trend from the user side. Developer Avid said he canceled a $200-a-month Claude subscription and moved to DeepSeek V4-Flash in place of Opus 5. The key is not a simple model swap but a setup that uses an open-source repo called "model-router" to attach DeepSeek V4-Flash to Codex subagents. In his words, "you install it and flip a toggle, and it takes 3 minutes."
No. 1 in Cline, and Switching Takes 3 Minutes?
1. It was perceived performance, not price-performance, that split the field. Cline usage is the result of models being chosen repeatedly in real coding work, not benchmarks. Users coming back is a sign it holds up in practice.
2. Subagent combinations lowered the barrier to entry. With tools like model-router, a setup that keeps the main model as is and spreads only the expensive parts to a cheaper model is possible with a 3-minute setup. This is a choice that changes the cost structure itself, not just a model swap.
3. $200 a month is a symbolic number. Canceling the top tier of the pro plans means cheaper models are starting to be seen not as "budget alternatives" but as "good enough replacements."
These Numbers Don't Speak for the Whole Market
Cline's usage data is a ranking among Cline users, not the share of the whole coding agent market. Avid's account is also one person's qualitative experience, and it is not evidence that the model-router setup delivers Opus 5-class results on every task. Still, it is clear that the official usage metric and real-user accounts are pointing in the same direction: a move to cheaper models.
Cost Structure Split First; Workflow Is Next
The standard for choosing a model is shifting from "which model is the smartest?" to "which combination holds up in real work?" The overlap between Cline's rankings and developers' reports of canceling and switching is the moment that shift starts to show up in the numbers.
Related: Cline's official X post · Avid's X post