Sierra Research

τ²-bench Retail

Share of tasks an agent completes on one attempt, against a simulated user who holds information it must ask for.

In index
Agentic
Source ↗

Models23

Top result85.3%Gemini 3.0 Pro

Top-3 spread1.1 pts

Updated30 Apr 2026

Result and price

Results

23 results

#ModelResultCost

01

Gemini 3.0 ProSierra85.3%—

02

Qwen3.5 397B A17Benabled effort · Sierra84.4%—

03

Nemotron Orchestrator 8BNVIDIA84.2%—

04UNRAFT 30B A3BNortheastern University HAI Lab82.5%$0

05

Claude Opus 4.1Anthropic82.4%—

06

GPT-5.2high effort · Sierra81.6%$0.072

07

GPT-5Sierra81.6%$0.106

08

Claude Opus 4Anthropic81.4%—

09

DeepSeek V3.2DeepSeek81.1%—

10

Claude Sonnet 4Anthropic80.5%—

11

Claude Opus 4.5high effort · Sierra79.6%$0.387

12

Qwen3 Max ThinkingQwen79.4%—

13

Claude Sonnet 4.5enabled effort · Sierra79.3%$0.255

14

Gemini 3 Flashhigh effort · Sierra76.8%—

15

Gemini 3 Prohigh effort · Sierra75.9%—

16

GPT-4.1Sierra74.0%—

17

o3OpenAI73.9%—

18

GLM-5enabled effort · Sierra73.7%—

19

Qwen3 MaxAlibaba72.2%—

20

Claude 3.7 SonnetSierra72.1%—

21MAKimi K2Moonshot AI70.6%—

22

o4-mini (high)Sierra68.3%—

23

GPT-4.1 miniSierra61.4%—