Vibe Coding Tools in Real Projects: What 883 Stacks Reveal (2026)
Of 883 public VCL projects with recorded tool data, 50.7% reported using multiple AI tools. See the three combinations builders reported most often.
Updated October 23, 2026. The tables now use dataset v9, with data through September 30. Sample sizes and pairings replace the earlier v7 figures; this is an edition update, not a measurement of changes within the same projects.
A product can have one name and several AI tools behind it.
In VCL’s data, just over half of projects with recorded tool data reported using multiple AI tools. That makes a ranking of individual tools only part of the story.
We reviewed 1,337 approved, public VCL projects. Of those, 883 recorded at least one AI tool: 66.0% of the sample. The other projects have missing tool data; that does not mean they used no AI tools.
Claude and Claude Code are counted together, as are ChatGPT and Codex. Each combined entry counts once; the data doesn’t distinguish which product was used.

Download the chart · Download the source CSV
What the 883 projects reported
| Measure | Result |
|---|---|
| Most-selected AI tool | Anthropic (Claude / Claude Code): 444 projects (50.3%) |
| One recorded AI tool | 435 projects (49.3%) |
| Two or more recorded AI tools | 448 projects (50.7%) |
The three most common combinations
Among the 448 projects reporting at least two AI tools, these pairs appeared most often:
| Reported combination | Projects | Share of 448 multi-tool projects |
|---|---|---|
| OpenAI (ChatGPT / Codex) + Anthropic (Claude / Claude Code) | 174 | 38.8% |
| Anthropic (Claude / Claude Code) + Cursor | 122 | 27.2% |
| OpenAI (ChatGPT / Codex) + Cursor | 122 | 27.2% |
The leading pair combines OpenAI and Anthropic tools. Cursor appears in both pairs tied for second.
These are overlapping pairs, not three exclusive stacks. A project listing all three tools contributes to all three rows.
How to use this when choosing your own tools
The split between single- and multi-tool projects is close to even. Treat the ranking as a starting point for comparison, not a shopping list.
Before adding another tool, ask:
- What job would it do that my current setup struggles with?
- Can I understand, test, and maintain what it produces?
If one tool already covers those jobs, this data offers no reason to add another.
How to read these numbers
These are builder- or submitter-reported selections on VCL, not a market-share estimate. The grouped entries don’t let us rank Claude Code alone or Codex alone. This snapshot does not show whether multi-tool use is increasing, how builders divided the work, or which combinations produce better products.
The category comparison shows how recorded tool choices differ between productivity apps, games, developer tools and social projects.
Data and source. State of Vibe Coding 2026, Dataset v9, cutoff October 1, 2026 at 00:00 UTC. The linked edition identifies the data and cutoff used for these figures.
The tool-stack chapter has the full ranking, pairings and downloadable tables.