Instruction FollowingNearing Saturation
Trust Score:
75

MT-Bench

80 curated multi-turn questions across 8 skills, each a two-turn conversation scored by a strong LLM judge on a 1–10 scale.

Launched: —Refresh: static
Status Assessment (nearing-saturation):

The 80 two-turn questions are fixed, public, and heavily exercised across research and leaderboards, so the test set is both widely memorized and increasingly saturated (top models score 8.5–9.0 on a 1–10 scale where 9 is a realistic ceiling). It remains a standard, still-discriminative LLM-judge eval, but the fixed set is the constraint: refreshed or held-out variants are the future direction.

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 Human Baseline (80.0%)
ModelScoreDateSource TypeProvenance
GPT-48.99%2023-06-09independentSource ↗

Human Baseline & Difficulty Horizon

Calibrated human reference points, specialist benchmarks, and ceiling thresholds.
Measured Human Score80.0%Domain Expert Baseline
Baseline Protocol & Interpretation

Human annotator agreement rate on multi-turn conversational dialogue quality.

Metric & Scoring Methodology

Verification protocols, aggregation formulas, and specialized metric variants.
Primary Metric:average judge score (1–10)
Scoring Engine:llm-judge

Dataset & Compute Cost

Evaluation volume, public availability, API pricing, and local hardware requirements.
Total Dataset Size80Annotated 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

Official Dataset & Benchmark Files:Download / View Dataset Repository ↗

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 Verifierllm-judge
Option AEleutherAI LM-Evaluation-Harness (Open-Weight Models)
lm_eval --model hf --model_args pretrained=<model_path> --tasks mt-bench --batch_size auto
Option BOpenCompass Evaluation Framework
opencompass --datasets mt-bench --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 MT-Bench, 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