← All benchmarks

Toloka Arena (v1)

Agents

An earlier task set of Toloka's agentic intelligence index (toloka.ai/arena), served as the page's v1 data: private, τ-bench-style multi-turn customer-service tool-use tasks with live databases and written business rules (the page described seven domains on September 9, 2026), scored by the composite pass^5, the probability that all five independent trials of a task succeed, at reasoning medium, 16,384 max tokens and temperature 0.6 unless a label states otherwise. Its cost data imply about 817 tasks; the count is not published. The page replaced it with a refreshed, smaller task set (toloka-arena-v3) by October 12, 2026. These results, last modified on September 23, are still served but no longer loaded by the page, and are kept as a separate historical benchmark.

Top models (higher is better)