Previous edition: Token-weighted tool_calls share per provider, April 2026

Tool-call share of output tokens and requests, October 2026

Tool-call share of output tokens and requests, October 2026

The same finish_reason mix, but weighted by output tokens instead of request count. Switch tabs to see how the picture changes.

By output tokensBy request countGap: token - request

Melious

66.4%

Melious

By output tokens66.4%By request count66.2%Gap (token - request)+0.2%

agentic and chat completions emit similar token volumes.

Lyceum

52.4%

Lyceum

By output tokens52.4%By request count50.6%Gap (token - request)+1.8%

agentic completions emit more tokens.

Vertex (Claude)

46.6%

Vertex (Claude)

By output tokens46.6%By request count36.0%Gap (token - request)+10.6%

agentic completions emit more tokens.

Novita

42.2%

Novita

By output tokens42.2%By request count9.4%Gap (token - request)+32.8%

agentic completions emit more tokens.

Sference

38.0%

Sference

By output tokens38.0%By request count38.0%Gap (token - request)0

agentic and chat completions emit similar token volumes.

Fireworks

32.6%

Fireworks

By output tokens32.6%By request count40.7%Gap (token - request)-8.1%

agentic completions are tighter.

Azure

31.7%

Azure

By output tokens31.7%By request count31.5%Gap (token - request)+0.2%

agentic and chat completions emit similar token volumes.

Bedrock

26.1%

Bedrock

By output tokens26.1%By request count20.2%Gap (token - request)+5.8%

agentic completions emit more tokens.

Mistral

24.3%

Mistral

By output tokens24.3%By request count26.1%Gap (token - request)-1.8%

agentic completions are tighter.

Vertex (Gemini)

23.5%

Vertex (Gemini)

By output tokens23.5%By request count18.3%Gap (token - request)+5.2%

agentic completions emit more tokens.

Anthropic

23.4%

Anthropic

By output tokens23.4%By request count22.7%Gap (token - request)+0.7%

agentic and chat completions emit similar token volumes.

OpenAI

21.4%

OpenAI

By output tokens21.4%By request count13.7%Gap (token - request)+7.7%

agentic completions emit more tokens.

DeepSeek

18.9%

DeepSeek

By output tokens18.9%By request count31.8%Gap (token - request)-12.9%

agentic completions are tighter.

TensorX

18.0%

TensorX

By output tokens18.0%By request count13.8%Gap (token - request)+4.3%

agentic completions emit more tokens.

Nebius

14.0%

Nebius

By output tokens14.0%By request count7.2%Gap (token - request)+6.7%

agentic completions emit more tokens.

DeepInfra

3.1%

DeepInfra

By output tokens3.1%By request count10.4%Gap (token - request)-7.4%

agentic completions are tighter.

How much provider traffic is tool calling? In October 2026 Melious had the highest tool-call share at 66.4% of output tokens and 66.2% of requests. Across the busiest providers, tool calls made up between 7.2% and 66.2% of successful requests.

Why it mattersTool calls are the core loop of agents. Comparing token share with request share shows whether tool-calling turns are longer or shorter than the average response on that provider.

Requesty (2026). Tool-call share of output tokens and requests, October 2026. Requesty Data. https://www.requesty.ai/data/tool-call-token-share-october-2026

@misc{requesty_tool_call_token_share_october_2026,
  author       = {{Requesty}},
  title        = {Tool-call share of output tokens and requests, October 2026},
  year         = {2026},
  howpublished = {\url{https://www.requesty.ai/data/tool-call-token-share-october-2026}},
  note         = {Requesty Data}
}