Tools & AgentsShunyu Yao et al. (Sierra)Shunyu Yao

τ-bench published

τ-bench published — a tool-agent benchmark in which the agent interacts with a user simulator while following domain rules

Why it matters

Measures policy compliance over multiple turns rather than one-off tool calls

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2406.12045

Added in Week 35, 2026 · See that week

Same week

LongRAG publishedPaperAnthropic releases Claude 3.5 SonnetOfficialSafe Superintelligence Inc. foundedOfficialMIPRO publishedPaper

Nearby on the same axis

Anthropic makes Claude tool use generally available (GA)OfficialSWE-agentPaperGitHub launches Copilot Workspace technical previewOfficialControversy spreads over whether Devin's Upwork demo holds upPress