AUSTRALIAAI MONITOR

Australian model map

Models from Australian developers and research collaborations involving Australian institutions. Explore the timeline, disclosed sizes and external dependencies.

25 of 25 model records · 17 multinational collaborations

Australian affiliation does not establish ownership or training location. Access labels are historical source observations. The inventory includes specialised and older research models alongside commercial systems.

8 plotted / 25 filtered models
Model publication date against parameter count, logarithmic scale100K1M10M100M1B10B100B20162017201820192020202120222023202420252026Publication yearReported total parameters · logarithmic scaleLumina-Image-2.0 · 2025-03-27 · 2.6B23CPDiffusion · 2024-09-10 · 4M22StarCoder · 2023-05-09 · 15.5B14Vega v2 · 2022-12-04 · 6B13bRSM + cache · 2019-12-02 · 2.55M12dense-IndRNN+dynamic eval · 2019-10-11 · 44.1M11EN^2AS with performance reward · 2019-07-22 · 23M10SPIDER2 · 2016-10-28 · 409.5K2
Australian organisation attributionMultinational collaborationEqual-sized markers · select for evidence
Model 23

Lumina-Image-2.0

Shanghai AI Lab,University of Sydney,Chinese University of Hong Kong (CUHK),Shanghai Jiao Tong University,Krea AI

Publication
2025-03-27
Parameters
2.6B
Base / component
Not established
Evidence and record details

Organisation-country attribution is not evidence of Australian ownership, training location or exclusive authorship. Historical model observation; current availability has not been re-audited. These records are not a census of Australian foundation models.

Dated Epoch research observation; source access terms are retained below.

Source snapshot retrieved 2026-09-20

Publication date
2025-03-27
Domain
Image generation
Task
Image generation,Text-to-image
Organization
Shanghai AI Lab,University of Sydney,Chinese University of Hong Kong (CUHK),Shanghai Jiao Tong University,Krea AI
Country (of organization)
China,Australia,Hong Kong,China,United States of America
Parameters
2600000000.0
Parameters notes
2.6B
Training compute (FLOP)
4.7794406e+21
Training compute estimation method
Hardware
Training compute notes
312000000000000 FLOP / GPU / sec [A100 reported, bf16 assumed] * 14184 GPU-hours [see training time notes] * 3600 sec / hour * 0.3 [assumed utilization] = 4.7794406e+21 FLOP
Training hardware
NVIDIA A100
Model accessibility
Open weights (unrestricted)
Open model weights?
Yes
Confidence
Confident
Last modified
2025-11-28 11:52:13+00:00
Epoch evidence ↗

Organisation-country attribution is not evidence of Australian ownership, training location or exclusive authorship.

Historical model observation; current availability has not been re-audited. These records are not a census of Australian foundation models.

Epoch AI — Lumina-Image-2.0 ↗Lumina-Image 2.0: A Unified and Efficient Image Generative Framework ↗Registry record →

Model key

How to read this graph · missing data and lineage

The vertical axis shows reported total parameters on a logarithmic scale. Model size does not establish performance or Australian ownership. Numbers match the model key and remain stable when filters change. Date filters limit the plotted points.

17 filtered records lack usable size/date evidence or have an unresolved source conflict. No values are imputed. The available base-model links refer to external models or components, so they appear in Dependencies; no Australian-to-Australian lineage arrows are inferred.