- Cover UI identifiers in the acronym guardrail
- Preserve FIG thumbnail and metadata values through archive parsing
- Use Vue-compatible acronym prop attributes
- Rename first-party API, RPC, JSON, CORS, SVG, JSX, and related identifiers to preserve acronym casing
- Keep upstream and serialized boundary names unchanged
- Add a lint guardrail and migration notes for exported APIs
- Separate model, connection, and advanced settings
- Resolve capabilities and output limits from provider catalogs
- Keep custom model compatibility declarations explicit
- Register invalidation immediately and release subscriptions on every failure path\n- Bound worker requests and settle parse and population errors consistently\n- Clear stale page loading state and retain current instance override imports
- Populate subsequent lazy pages in a development-only persistent worker
- Journal created, updated, and deleted graph state as field deltas
- Invalidate stale worker replicas after authoritative graph mutations
- Fall back to synchronous population when the worker is unavailable
- Ignore undefined values before cache invalidation and preview mutation\n- Treat text direction as a glyph-shaping input\n- Cover the corrected invalidation contract
Keep authoritative derived glyphs through layout, alignment, fill, and decoration changes while still invalidating them for actual shaping edits.
Add scene-graph invalidation coverage and a Gold Preview fixture regression test.
Co-authored-by: Joseph Cumines <joeycumines@gmail.com>
- Warn once with the rejected surface dimensions and original CanvasKit error
- Disable repeated cache allocation attempts until the renderer surface is replaced
- Keep direct scene rendering as the visible fallback
- Scale retained margins to a fixed device-pixel budget on large viewports
- Fall back to direct scene rendering when CanvasKit rejects an offscreen surface
- Cover the reported wide HiDPI geometry without coupling rendering to a WebGL context
- Traverse fixed wrappers only to reach descendants with SCALE constraints
- Preserve fixed siblings and use component geometry as the scale basis
- Cover target-aspect icons without changing the Preline geometry baseline
- Use preserved targetAspectRatio metadata as the instance scale basis
- Scale nested geometry through fixed wrapper layers
- Cover proportion-constrained logo imports with a regression test
- Scope recipes to OpenAI-compatible endpoints and enforce CORS status and header values
- Record unavailable ID coverage, output budgets, stream HTTP status, finish reasons, and zero-call failures
- Repair the repository README link to the published compatibility reference
- Recompute Yoga positions after shaped text changes effective geometry
- Resolve inherited stretch and skip hidden or absolute children
- Share and normalize Plugin API transform matrices
- Centralize design-profile eligibility for the selector, role assignment, fallback, and tests
- Keep built-in provider model selection intact while switching saved custom profiles
- Add localized accessible labeling and constrain long profile names
The chat model chip is a dropdown for built-in providers but a static label
for OpenAI-compatible and custom-model setups, so anyone using a custom
endpoint had to open Settings to change models -- even with several models
already configured.
Add a profile switcher that reassigns the design role from the chat panel.
It lists saved profiles that can use tools, shows the profile name rather
than the raw model ID, and badges the vision-capable ones. It writes the
same assignments.design field the Settings dropdown does, so the two stay in
sync.
Only renders when more than one tools-capable profile exists; a single
profile keeps the existing label. Built-in providers keep ProviderModelSelect
unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Return rotation and reflection matrices for relative and absolute transforms
- Preserve authoritative imported FIG transforms until geometry is edited
- Cover nested rotation and reflected vector behavior
- Reconcile stale generated text widths with CanvasKit measurements
- Propagate effective widths through HUG containers while preserving fixed geometry
- Cover component-lineage and stretched-text safeguards
- Publish the compatibility reference in the VitePress site and link localized AI guides to the canonical English page
- Correct current Anthropic and credential-storage behavior
- Make CORS, streaming, pricing, context, and reasoning measurements reproducible and security-safe
The probe behind this page used max_tokens 2000, which starves reasoning
models: they spend the whole budget thinking and emit no tool calls, with
finish_reason length and no error. That looks identical to a model that
cannot call tools, and two models were wrongly marked "avoid" because of it.
Re-ran every Scaleway model at 16384. All five are clean -- qwen3.6-35b-a3b
goes from 1/3 to 3/3, mistral-small-3.2-24b from 2/3 to 3/3. Tool calling no
longer separates them, so the recommendation now follows price, vision and
context: mistral-small-3.2-24b is cheapest with vision, gemma-4-26b-a4b-it
for double the context.
Document the failure mode, the reasoning_effort lever (only "none" had any
effect on Scaleway; "low" did nothing), and ask contributors to record the
token budget alongside results. Nebius and TensorX rows are marked as
measured at 2000 so they are not read as comparable.
Refs #454
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Add measured Vision and context-window columns alongside the tool-calling
metrics, and four more Scaleway models. gemma-4-26b-a4b-it becomes the
recommendation: the only model tested that is clean on cost, tool calling
and vision at once, with 2x the context of gpt-oss-120b.
Document models.dev as a prefill source for new rows, with the caveat that
it describes nominal capabilities rather than deployment behaviour. It was
wrong on two of seven models checked here -- it marks scaleway/gpt-oss-120b
as vision-capable when the model refuses images, marks
scaleway/mistral-small-3.2-24b as text-only when it reads them, and reports
tool_call: true for models that emit no tool calls in practice.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
