max_context Subscribe

MODELS + MODEL LAB

What we track. What we test.

Every field has a source. Unknown stays 404. Test scores appear only after a real run.

ModelLabLicenceAccessEUOur tests
DeepSeek-V4.1-FlashDeepSeekMITOpen weights (Hugging Face)not checked404: no runs
Gemini 4 ArgonGoogle DeepMind404404not checked404: no runs
FLUX 3 KleinBlack Forest Labsother (read it)Open weights (Hugging Face)not checked404: no runs
MOSS-Image404Apache-2.0Open weights (Hugging Face)not checked404: no runs
Qwen-Image-2.1-TurboQwenother (read it)Open weights (Hugging Face)not checked404: no runs
LAB

Model Lab

Radar, skyline, scatter, compare. Real results only; flip EXAMPLE to see how it will look.

5models tracked
5tests planned
0runs logged
404top score
5not run yet: 404
404No test runs yet.We never guess a score. The first runs come once the test set is locked.
REAL DATA: 5 models tracked, 0 runsdrag to orbit / click a node for its page
3D needs WebGL. Every model's page lists what we know (and the 404s).
Test results
ModelKitchen-table tasksFine printRead the receiptCount the fingersSays 404
DeepSeek-V4.1-Flashnot run yetnot run yetnot run yetnot run yetnot run yet
Gemini 4 Argonnot run yetnot run yetnot run yetnot run yetnot run yet
FLUX 3 Kleinnot run yetnot run yetnot run yetnot run yetnot run yet
MOSS-Imagenot run yetnot run yetnot run yetnot run yetnot run yet
Qwen-Image-2.1-Turbonot run yetnot run yetnot run yetnot run yetnot run yet

THE TESTS (PLANNED, NOT RUN YET)

  • Kitchen-table tasksEveryday jobs: summarise a lease, plan a bus route, split a bill fairly.planned
  • Fine printFind the catch in real terms-of-service and pricing pages.planned
  • Read the receiptPhotos of receipts and labels. Get the numbers right.planned
  • Count the fingersImage models: prompts with hands, text and clocks.planned
  • Says 404Questions with no known answer. Does it say "I don't know"?planned