tensor.news
Integrity IndexModelsCompareLabsPapersMethodologyAbout
tensor.news

© 2026

Evidence

  • Integrity Index
  • Model Claim Ledger
  • Compare

Entities

  • Models
  • Benchmarks
  • Labs
  • Papers

Method

  • Methodology
  • About
  • Privacy

Machine

  • Feed (Atom)
  • llms.txt
  • llms-full.txt
  • Sitemap

tensor.news grades benchmarks and tags model claims by evidence status. Every fact carries a source record and provenance. Scores are framed as task performance under a disclosed harness, never as deployed capability.

← Integrity Index

ScienceQA

ScienceQA

Rank #42 of 51 · 6 models scored · top score 84.67

CINTEGRITY 68 / 100

unknown

Strongest on discrimination, weakest on contamination resistance. Caveat: scores come from mixed harnesses, so a head-to-head is not apples-to-apples.

mixed harness, not directly comparablepublic test set

Headroom to ceiling

15.3 headroomtop 84.7ceiling 100

Score distribution

median 24.4GPT-4o (May 2024) 84.7125088

State of the art over time

12508884.720232024

Integrity breakdown

Weighted composite. Each signal is computed from the source-backed leaderboard below.

Discrimination30%100

Does it still separate models?

Saturation headroom30%86

How far from ceiling / clustered at the top?

Contamination resistance15%0

Public vs held-out; training-leak risk.

Harness comparability15%50

Apples-to-apples, or mixed / vendor-optimized?

Freshness10%44

How old is the benchmark?

Leaderboard

#ModelScoreEvidenceMeasured
1GPT-4o (May 2024)84.67self-reported· optimized2024-05-13
2Claude 3 Haiku62.67self-reported· optimized2024-03-04
3Llama 2-13B41.04unverified· optimized2023-07-18
4LLaMA-13B24.44unverified· optimized2023-02-27
5Llama 2-7B24.11unverified· optimized2023-07-18
6LLaMA-7B14.92unverified· optimized2023-02-24

At a glance

Models scored
6
Top model
GPT-4o (May 2024)
Top score
84.67
Score spread
69.75
Headroom
15.33

Benchmark context

Task type
multimodal
Metric
normalized performance in Epoch CSV, rendered as accuracy (%)
Score ceiling
100
Human baseline
unknown
Test-set privacy
unknown
Construction
unknown
Key assumptions
unknown
Limitations
unknown
Contamination history
unknown
Harness notes
Epoch CSV reports model, model version, performance, optimized flag, score source, benchmark release date, and model date.
Authors
unknown
Released
2022-09-20
License
CC BY 4.0
Source
epoch.ai/data/eci_benchmarks.csv

Provenance

6 source records · scores from epoch_benchmarks. Every fact on this page is evidence-linked.

Follow the record

Integrity neighbors

  • MATH level 5B · 73
  • BBHB · 71
  • GSM8KC · 69
  • HellaSwagC · 67
  • MMLUC · 67
  • ARC AI2C · 66

Top models here

  • GPT-4o (May 2024)84.67
  • Claude 3 Haiku62.67
  • Llama 2-13B41.04
  • LLaMA-13B24.44
  • Llama 2-7B24.11

Head-to-head

  • GPT-4o (May 2024) vs Claude 3 Haiku
  • GPT-4o (May 2024) vs Llama 2-13B
  • Claude 3 Haiku vs Llama 2-13B