Is τ²-bench still meaningful?
Status:Near ceiling Checked against the report, 5 November 2026
Verdict
Fading. Top scores are high, and the newer τ³-Banking spreads models more.
Why near ceiling: The chapter rates it Near ceiling: the top score is 87.9%, and telecom sits near 99% (secondary).[1][1] τ-bench leaderboard Sierra Primary
What “Near ceiling” meansKnown problems
What to look at instead
- τ³-Banking Status:New / unproven Promising. The best agent scores 55.2%, but outside review is thin.
Where it appears in the report
Chapter 1, Benchmarks
- Table 7. General and computer-use agent benchmarks, November 2026 in “Go deeper: General and computer-use agent benchmarks (18)”
Chapter 2, Evaluation science
- Error bars and adaptive tests make a score trustworthy at lower cost
- Agent scores now depend on how you count cheating and spend tokens
Chapter 3, Tools and technology
Sources
Copied from the chapter source lists. Each number links back to the source in its chapter.
- [1] τ-bench leaderboard, Sierra. It gives τ²-bench, τ³-Banking and τ-Voice scores. Primary taubench.com Chapter 1, source 22
- [2] Agentic Benchmark Checklist, arXiv 2507.02825. It finds scoring flaws in agent benchmarks. Primary arxiv.org Chapter 1, source 173