A month-by-month grid of major model families published from July 2025 through July 2026 across language, image, video, vision, audio, embeddings and 3D. Parameter variants, quantizations, FP8/GGUF builds and routine checkpoints are collapsed so the density reflects families and meaningful releases, not repository duplication.
—release cards
—labs represented
13month columns
—model types
How to read the cards
The large number is the total parameter count of the representative flagship or main open checkpoint for that family. For MoE models, active-parameter counts are intentionally omitted so sizes stay visually comparable.
ReasoningMultimodalCodingAgenticMoE
The flag identifies the lab’s home country. The small label at the top-right of each card gives its primary model type; capability chips below can still mark reasoning, multimodal, coding, ASR, TTS, generation, retrieval and related traits.
Source & caveat
This is a curated visualization layer built around official Hugging Face model pages and collections from major labs. It is intentionally not a literal export of every repository: small individual publishers, quantizations, mirrors and routine checkpoint variants are excluded.
Release dating can differ between repository creation, announcement, and later model-card updates. This grid uses the first public release month for the named family/release where possible.