Tools models call · September 2026
4.3% of chat requests in September 2026 returned at least one tool call. The most widely called tool name was "read", seen at 37.4% of organizations using tools.
Tool names that models called, ranked by how many tool-using organizations saw each one, plus the overall share of requests with a tool call.
Data through September 30, 2026Updated November 11, 2026
This is the permanent snapshot for September 2026. See the latest month (October 2026)
Share of tool-using organizations, September 2026
Source: Requesty production gateway- read37.4%
- bash34.5%
- edit29.7%
- write28.5%
- read_file22.9%
- grep21.6%
- web_search21.4%
- glob21.0%
- skill16.2%
- write_file14.0%
- todowrite13.5%
- search_files13.4%
- terminal12.9%
- webfetch12.0%
- execute_code11.5%
- skill_view11.2%
- memory10.0%
- skill_manage9.7%
- patch9.5%
- task8.7%
Key findings
Share of tool-using organizations by month
What changed since August 2026
- bash38.6% → 34.5% (-4.1 pp)
- read_file26.7% → 22.9% (-3.8 pp)
- glob24.3% → 21.0% (-3.3 pp)
- write31.8% → 28.5% (-3.3 pp)
- grep24.6% → 21.6% (-3.0 pp)
What to do about it
- Keep tool definitions stable between turns so they stay in the prompt cache.
- Pick models with strong tool calling for agent work and check their error rate on /rankings.
Every tool call is logged with its arguments and latency in the Requesty dashboard.
Read about request logsData
| Tool name | Share of tool-using organizations | Share of tool-call requests | Organizations (band) |
|---|---|---|---|
| read | 37.44% | 9.91% | 100-499 |
| bash | 34.51% | 21.09% | 100-499 |
| edit | 29.66% | 3.70% | 100-499 |
| write | 28.49% | 1.58% | 100-499 |
| read_file | 22.91% | 1.24% | 100-499 |
| grep | 21.59% | 2.96% | 100-499 |
| web_search | 21.44% | 2.88% | 100-499 |
| glob | 21.00% | 0.80% | 100-499 |
| skill | 16.15% | 0.36% | 100-499 |
| write_file | 13.95% | 0.33% | 25-99 |
| todowrite | 13.51% | 0.47% | 25-99 |
| search_files | 13.36% | 0.21% | 25-99 |
| terminal | 12.92% | 2.42% | 25-99 |
| webfetch | 12.04% | 0.11% | 25-99 |
| execute_code | 11.45% | 0.62% | 25-99 |
| skill_view | 11.16% | 0.24% | 25-99 |
| memory | 9.99% | 0.07% | 25-99 |
| skill_manage | 9.69% | 0.18% | 25-99 |
| patch | 9.54% | 0.24% | 25-99 |
| task | 8.66% | 0.27% | 25-99 |
| web_extract | 7.93% | 0.07% | 25-99 |
| skills_list | 7.78% | 0.03% | 25-99 |
| question | 7.78% | 0.05% | 25-99 |
| websearch | 7.64% | 0.07% | 25-99 |
| agent | 7.64% | 0.09% | 25-99 |
| apply_patch | 6.46% | 0.85% | 25-99 |
| askuserquestion | 6.02% | 0.05% | 25-99 |
| ask_user | 6.02% | 0.10% | 25-99 |
| process | 5.73% | 0.08% | 25-99 |
| list_files | 5.73% | 0.04% | 25-99 |
| search_web | 5.43% | 0.07% | 25-99 |
| clarify | 5.43% | 0.02% | 25-99 |
| exec | 5.29% | 0.44% | 25-99 |
| delegate_task | 5.29% | 0.02% | 25-99 |
| browser_exec | 5.14% | 0.03% | 25-99 |
| shell | 5.14% | 0.29% | 25-99 |
| tool_call | 5.14% | 0.13% | 25-99 |
| read_files | 5.14% | 0.19% | 25-99 |
| web_fetch | 5.14% | 0.05% | 25-99 |
39 rows published. Rows below the floors are withheld.
Questions this answers
- What tools do AI agents call most?
- 4.3% of chat requests in September 2026 returned at least one tool call. The most widely called tool name was "read", seen at 37.4% of organizations using tools.
- What share of LLM requests use tool calling?
- 4.3% of chat requests in September 2026 returned at least one tool call. The most widely called tool name was "read", seen at 37.4% of organizations using tools.
- Which agent tools are most common: read, bash, edit or web search?
- 4.3% of chat requests in September 2026 returned at least one tool call. The most widely called tool name was "read", seen at 37.4% of organizations using tools.
Method and definitions
- Tool names come from the tool calls in model responses, lowercased. A tool counts once per request even if it is called several times.
- A tool name needs 10 or more organizations before it is listed, on top of the standard floors, so names specific to one company never appear.
- Tool-call data is recorded from July 2026, so this series starts then.
- A row is published only when it has at least 20 requests, at least 5 distinct organizations, and no single organization above 80% of its traffic. These are the same floors /rankings uses. Rows below the floors are withheld, not estimated.
- Only shares, rates, medians and organization bands are published. Absolute request, token and spend totals stay private.
- Reach:
- Organizations whose requests called this tool, divided by organizations with any tool call in the month.
- Tool-call request:
- A chat request whose response contained at least one tool call.
Caveats
- Different agents name the same tool differently (read vs read_file, bash vs shell vs terminal). Names are not merged.
- Only tool calls the model returned are counted, not tools that were offered but unused.
Revision history and other months
- October 2026first published later
- September 2026 (this page)first published 2026-11-11
- August 2026first published 2026-11-11
- July 2026first published 2026-11-11
Cite and reuse
Free to reuse under CC BY 4.0. Credit "Requesty Data" and link to this page. Cite the month URL so the numbers you quote stay fixed.
APA
Click to copy
BibTeX
Click to copy
CSV
Click to copy