- Who it helps
- Researchers and developers evaluating tool-using agents.
- Availability
- November 4 preprint, project explanation and public diagnostic repository are available. Full training reproduction was not verified.
Measure both attempts and turns
Pass@(K,T) asks whether any of K attempts succeeds within T interaction turns. Across four benchmarks and two Qwen model sizes, the authors find stronger error recovery without consistent improvement in every individual decision.
The budget is part of the test
The proposed training method samples different turn budgets. Reported gains under tighter limits suggest evaluations should match deployment constraints; the evidence remains limited to the tested tasks and configurations.
BudgetMix tests agent training under tighter turn limits ↗BudgetMix project explanation ↗Rethinking Agentic RL diagnostics ↗BudgetMix paper discussion ↗
What you can do
- Read the linked primary evidence and its access limits before planning a trial. Open reference ↗
Where the evidence stops
No experiments reproduced. Results with an announced turn cap may differ from an unexpected timeout or spending cutoff.
Sources read. Sources read November 5, 2026 UTC. No third-party project, model inference or complete product workflow was executed.
Sources & further reading
Paper · paper
BudgetMix tests agent training under tighter turn limits ↗
Primary page read November 5. Claims were checked against the page; the workflow was not executed.
Kejian Zhu and collaborators · page checked 2026-11-05
GitHub · implementation
Rethinking Agentic RL diagnostics ↗
Primary page read November 5. Claims were checked against the page; the workflow was not executed.
Kejian Zhu and collaborators · page checked 2026-11-05
16 stars · observed 2026-11-05 13:19 UTC
More sources reviewed (2)
Web · paper
BudgetMix project explanation ↗
Primary page read November 5. Claims were checked against the page; the workflow was not executed.
Kejian Zhu and collaborators · page checked 2026-11-05
Hugging Face · discussion
BudgetMix paper discussion ↗
Primary page read November 5. Claims were checked against the page; the workflow was not executed.
Hugging Face contributors · page checked 2026-11-05
Coverage note: Original YouTube demonstrations and independent cross-platform attention were not established. No end-to-end workflow was executed.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
Agent-training gains can shrink when an agent has fewer turns to finish. BudgetMix varies those limits during training.