mirror of
https://github.com/Twigpine/model-releases.git
synced 2026-11-07 11:45:57 -10:00
OpenClaude
7c18877ef5
Move from prose-in-markdown to one YAML file per model under models/, validated against a JSON Schema in CI. journal/*.md and INDEX.md are now generated from the YAML by scripts/build.py — single source of truth. Adds 11 new entries covering the April 2026 releases listed in the @RoundtableSpace tweet (Claude Opus 4.7, DeepSeek V4, Xiaomi MiMi V2.5, Qwen3.6 variants, GLM 5.1, Muse Spark, Grok 4.3 Beta, Kimi K2.6) as stubs — enough to make them discoverable, sparse where sources weren't available. Also: CONTRIBUTING.md, PR template, GitHub Actions workflow that fails if generated files drift from the YAML.
Model Release Journal
An open, community-curated record of AI model releases.
- Structured data: one YAML file per model under
models/. - Validated: every entry is checked against
schema/model.schema.jsonin CI. - Generated views:
INDEX.md(all models, sortable by date or provider) andjournal/(grouped by month) are rebuilt from the YAML — do not edit them by hand.
See INDEX.md for the current list.
Why
Model releases happen constantly. Pricing shifts, context windows expand, naming conventions drift. Storing this as structured data — not prose — makes it diffable, queryable, and durable.
Schema
Each models/<model_id>.yaml entry has these fields (see schema/model.schema.json for the authoritative definition):
| Field | Required | Description |
|---|---|---|
model_id |
yes | Canonical slug, lowercase. Must match the filename stem. |
display_name |
no | Human-readable name. |
provider |
yes | Organization that serves the model. |
release_date |
yes | ISO 8601 date of public availability. |
context_window |
no | Input tokens (integer) or null. |
output_tokens_max |
no | Max output tokens per response. |
modalities |
no | Subset of text, code, image, audio, video, pdf, embedding. |
status |
no | preview, stable (default), deprecated, retired. |
supersedes |
no | model_id of the predecessor, if tracked here. |
pricing |
no | { input_per_mtok, output_per_mtok, cached_input_per_mtok }. |
sources |
no | URLs of primary sources. |
tags |
no | Free-form labels. |
notes |
no | Short prose on what changed. |
Contributing
See CONTRIBUTING.md. Short version:
cp models/TEMPLATE.yaml models/<model_id>.yaml
# edit fields
python -m venv .venv && .venv/bin/pip install pyyaml jsonschema
.venv/bin/python scripts/validate.py
.venv/bin/python scripts/build.py
git add models/ journal/ INDEX.md
Repository layout
schema/model.schema.json # JSON Schema for entries
models/<model_id>.yaml # one file per model (source of truth)
models/TEMPLATE.yaml # copy this to start a new entry
scripts/validate.py # schema + sanity checks (CI enforces)
scripts/build.py # regenerates journal/ and INDEX.md
journal/YYYY-MM.md # generated monthly view
INDEX.md # generated global index
.github/workflows/ # CI
License
MIT