Agentic & Tool UseActive
Trust Score:
93

τ³-Banking

Banking customer-service agent benchmark from the τ-bench family, evaluated by Artificial Analysis.

Launched: —Refresh: static
Status Assessment (active):

Actively evaluated by Artificial Analysis on current frontier models; no saturation evidence recorded.

Performance Timeline

Longitudinal progression of model scores against human baselines.

Performance & Historical Trajectory

Empirical score progression across model release dates and evaluation rounds.

Independent Vendor-Reported
ModelScoreDateSource TypeProvenance
Llama 4 Maverick3.7%2026-11-15independentSource ↗
Qwen3.8 Max (0902)47.8%2026-11-15independentSource ↗
Motif 335.3%2026-11-15independentSource ↗
Granite 4.2 8B7.6%2026-11-15independentSource ↗
Llama 4 Scout3.3%2026-11-15independentSource ↗
Qwen3.7 Plus17.5%2026-11-15independentSource ↗
K2 Horizon 0.9B4.9%2026-11-15independentSource ↗
Qwen3.8-Flash-Next45.4%2026-11-15independentSource ↗
Ministral 3 8B3.7%2026-11-15independentSource ↗
Hy322.9%2026-11-15independentSource ↗
Qwen3.8 2.4T A95B49.1%2026-11-15independentSource ↗
Kimi K2.7 Code20.2%2026-11-15independentSource ↗
Magistral Medium 1.25.6%2026-11-15independentSource ↗
DeepSeek V4-Pro39.6%2026-11-15independentSource ↗
A.X-K216.1%2026-11-15independentSource ↗
Celeris-13.9%2026-11-15independentSource ↗
Ling 3.0 Tiny20.8%2026-11-15independentSource ↗
Solar Open2 250B21.6%2026-11-15independentSource ↗
Gemma 4 31B14.8%2026-11-15independentSource ↗
Qwen3 Coder Next5.4%2026-11-15independentSource ↗
KAT Coder Pro V24.5%2026-11-15independentSource ↗
Quasar 438B28.7%2026-11-15independentSource ↗
Agnes 2.5 Pro Beta35.7%2026-11-15independentSource ↗
Ring-2.6-1T17.9%2026-11-15independentSource ↗
Inkling 2 Small40.4%2026-11-15independentSource ↗
LFM2.5-2.6B7.2%2026-11-15independentSource ↗
Mistral Medium 3.515.1%2026-11-15independentSource ↗
DeepSeek V4 Flash Vision41%2026-11-15independentSource ↗
GLM 5.350.3%2026-11-15independentSource ↗
gpt-oss-120b12.8%2026-11-15independentSource ↗
Qwen3 Next 80B A3B (Reasoning)6.2%2026-11-15independentSource ↗
GPT-5.5 Instant11.8%2026-11-15independentSource ↗
G9v3-39A5B22.1%2026-11-15independentSource ↗
Gemini 3.1 Pro21.4%2026-11-15independentSource ↗
K2 Horizon 375B A23B34.2%2026-11-15independentSource ↗
Trinity Large Thinking5.8%2026-11-15independentSource ↗
Solar Pro 423.3%2026-11-15independentSource ↗
Qwen3.5 4B4.3%2026-11-15independentSource ↗
Command A+6%2026-11-15independentSource ↗
Nemotron 3.5 Lightning8.9%2026-11-15independentSource ↗
MiniMax M315.3%2026-11-15independentSource ↗
GPT-6 Astra41.4%2026-11-15independentSource ↗
North Mini Code6.4%2026-11-15independentSource ↗
GLM 5.3 Flash47.2%2026-11-15independentSource ↗
K-EXAONE 2.011.5%2026-11-15independentSource ↗
K2 Horizon 3.7B19.6%2026-11-15independentSource ↗
Mistral Small 44.9%2026-11-15independentSource ↗
K2 Horizon 7B27.6%2026-11-15independentSource ↗
Granite 4.2 30B14.4%2026-11-15independentSource ↗
Qwen3.5 397B A17B13.4%2026-11-15independentSource ↗
Inkling29.1%2026-11-15independentSource ↗
Granite 4.2 3B5.6%2026-11-15independentSource ↗
Gemini 3.5 Flash-Lite17.5%2026-11-15independentSource ↗
Apodex 1.125.2%2026-11-15independentSource ↗
MiniCPM5-2B20.8%2026-11-15independentSource ↗
Gemini 3.8 Flash44.9%2026-11-15independentSource ↗
Inkling Small18.8%2026-11-15independentSource ↗
Ministral 3 14B6.6%2026-11-15independentSource ↗
Magistral Small 1.24.5%2026-11-15independentSource ↗
Gemma 4 26B A4B12%2026-11-15independentSource ↗
Ling-3.0-flash-Fin38.6%2026-11-15independentSource ↗
Ministral 3 3B4.7%2026-11-15independentSource ↗
Qwen3.6 35B A3B5.4%2026-11-15independentSource ↗
Qwen3.5-122B-A10B10.3%2026-11-15independentSource ↗
LongCat 2.013.2%2026-11-15independentSource ↗
Nemotron 3 Ultra14.2%2026-11-15independentSource ↗
Muse Glimmer23.5%2026-11-15independentSource ↗
K2 Horizon MoVA 36B A4B32%2026-11-15independentSource ↗
Ling-3.0-flash-VL34.4%2026-11-15independentSource ↗
Nex-N2-Pro17.5%2026-11-15independentSource ↗
Qwen3.8 27B48%2026-11-15independentSource ↗
HyperNova 60B 26055.6%2026-11-15independentSource ↗

Human Baseline & Difficulty Horizon

Calibrated human reference points, specialist benchmarks, and ceiling thresholds.
Measured Human Scoren/aNot established
Baseline Protocol & Interpretation

No peer-reviewed human baseline has been published for this benchmark.

Metric & Scoring Methodology

Verification protocols, aggregation formulas, and specialized metric variants.
Primary Metric:task success (%)
Scoring Engine:composite

Dataset & Compute Cost

Evaluation volume, public availability, API pricing, and local hardware requirements.
Total Dataset Size—Annotated evaluation items
Public Test SetPublicOpenly mirrored on repositories
Access GatingOpen AccessUnrestricted download
Evaluation LicenseOpen AccessDataset usage and redistribution terms
Frontier API Compute Cost:

$5 – $20 USD for full benchmark evaluation run on frontier APIs.

Recommended Local GPU Setup:

1x NVIDIA RTX 4090 (24GB) or A100 (40GB/80GB) via vLLM / SGLang

How to Run & Reproduce

Standardized evaluation protocols, CLI commands, and reproducible runner templates.
Prompt Regimezero-shot
Reasoning Modedirect
Sampling Temp0
Pass@k Budgetk = 1
Tools & SandboxPure Text
Scoring Verifiercomposite
Option AEleutherAI LM-Evaluation-Harness (Open-Weight Models)
lm_eval --model hf --model_args pretrained=<model_path> --tasks tau3-banking --batch_size auto
Option BOpenCompass Evaluation Framework
opencompass --datasets tau3-banking --models <model_config>
Python APIDeterministic Inference Loop Snippet
# Standard API Evaluation Loop
from openai import OpenAI

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": prompt}],
    temperature=0.0,
)
Standardized Reporting Requirement:

When publishing results for τ³-Banking, always report the exact prompt template, few-shot exemplar ordering, sampling temperature (temperature=0), maximum reasoning budget tokens, and the precise timestamped model snapshot ID.

Contamination & Memorization Analysis

Audit of pretraining exposure risks, memorization vectors, and refresh policies.
Overall Contamination Risk:MEDIUM
Refresh Cadence:

Static fixed snapshot

Test Set Exposure:

Public on web / HuggingFace