Large Language Models

A comparative timeline of frontier & open-weight models · 2024–2026 · Hover dots for details

Tooltip – Model Info

Release date

Parameter count

Context window

Knowledge cutoff

AA Ratings (0–10 segments)

CD — Cognitive Depth

CO — Coding

AG — Agentic Reliability

SC — Scientific Reasoning

UX — UX & Empathy

Timeline Dots

Flagship (proprietary)

Release (proprietary)

Flagship (open weights)

Release (open weights)

Not yet released (leak / teaser)

Modality Badges

LLM Text / Language

VIS Image understanding

IMG Image generation

AUD Audio understanding

VID Video understanding

VGEN Video generation

Index Methodology

Cognitive Depth

Direct AA Intelligence Index score, normalised so the current best model = 10 segments.

AA Intelligence Index 100%

Coding

Weighted composite of two AA coding benchmarks, normalised to best model = 10 segments.

Terminal-Bench Hard 66.6%

SciCode 33.3%

Agentic Reliability

Weighted composite of two AA agentic benchmarks, normalised to best model = 10 segments.

GDPval 66.6%

τ²-Bench Telecom 33.3%

Scientific Reasoning

Weighted composite of three AA science benchmarks, normalised to best model = 10 segments.

HLE 50%

GPQA Diamond 25%

CritPt 25%

UX & Empathy

Weighted composite of four EQ-Bench 3 sub-scores, min-max normalised across all models to 1–10 segments.

EQ-B-3 Humanlike 30%

EQ-B-3 Social Dexterity 25%

EQ-B-3 Warmth 15%

EQ-B-3 Empathy 30%

Model Name

Model Name