Requesty
Data/Agentic workloads

Previous edition: finish_reason mix per model, April 2026

finish_reason mix per model, October 2026

finish_reason mix per model, October 2026

Successful completions only, normalized to 100%. Click a legend item to isolate.

Requesty
In October 2026 deepseek-v4.1-flash-1017 ended 73% of successful completions with tool_calls, the highest among the busiest models.Top models by usage; finish_reason was not populated before 2026. Source: Requesty production gateway.

Which models are used most for tool calling? In October 2026 deepseek-v4.1-flash-1017 ended 73% of successful completions with tool_calls, the highest among the busiest models. gpt-4.1-mini was at 0%, used almost entirely for plain completions.

Why it mattersModels with a high tool_calls share are the ones teams run inside agents. A low share points to chat, extraction or batch use.

Period
Oct 2026
Updated
November 11, 2026
ID
finish-reason-by-model-october-2026
§ 01

Key findings

  • 01deepseek-v4.1-flash-1017: 73% tool_calls.
  • 02gemini-3.8-flash: 63% tool_calls.
  • 03deepseek-v4.1-flash: 58% tool_calls.
  • 04glm-5.3-flash: 48% tool_calls.
  • 05gpt-6-luna: 27% tool_calls.
§ 01b

Questions this answers

  • Which LLM is used most for tool calling?
  • What is the finish_reason mix by model?
  • Which models are used for agents?
§ 02

Data

Modeltool_calls(percent)stop(percent)length(percent)
deepseek-v4.1-flash-101772.88%25.57%1.55%
gemini-3.8-flash63.13%34.99%1.88%
deepseek-v4.1-flash57.78%40.56%1.66%
glm-5.3-flash48.34%43.74%7.92%
gpt-6-luna27.17%72.30%0.53%
claude-opus-5-525.20%72.76%2.03%
gpt-5.6-luna24.76%74.59%0.65%
claude-sonnet-523.95%66.31%9.74%
deepseek-v4-flash-07318.22%90.29%1.49%
gemma-4-26B-A4B-it3.29%96.68%0.03%
gemini-3.5-flash-lite1.92%98.06%0.02%
gemma-4-31B-it0.98%98.98%0.04%
gemini-2.5-flash-lite0.46%99.32%0.22%
gpt-4.1-mini0.37%99.59%0.05%
§ 03

Cite as

APA
Click to copy
BibTeX
Click to copy
ID: finish-reason-by-model-october-2026·Updated November 11, 2026·Period Oct 2026