[diegosouzapw]

Closes #4073

Root cause (confirmed against the live models.dev API)

getResolvedModelCapabilities("mistral/pixtral-12b-latest").supportsVision resolved null — vision came only from the #4071 model-id heuristic, with attachment still null — even though Pixtral is multimodal.

models.dev does expose the data; it is not an upstream gap. It catalogs Pixtral 12B under the short id pixtral-12b:

mistral / pixtral-12b          | attachment=true  | modalities.input=["text","image"]
mistral / pixtral-large-latest | attachment=true  | modalities.input=["text","image"]
mistral / mistral-medium-latest| attachment=true  | modalities.input=["text","image"]

Requests use the Mistral API alias pixtral-12b-latest. getSyncedCapabilityForResolved tried the exact id, the raw id and the static-spec canonical id — all of which miss the short form — so it fell through to the heuristic (steps 1–4 of resolveVisionCapability all missed). Models whose -latest id is stored verbatim (pixtral-large-latest, mistral-medium-latest) already resolved correctly; only the short-formed pixtral-12b was broken.

Fix

getSyncedCapabilityForResolved gains a last-resort fallback: after exact / raw / canonical all miss, retry the synced lookup with a trailing -latest stripped (pixtral-12b-latest → pixtral-12b). Synced metadata (attachment / image modalities) then wins over the heuristic. Because it only fires after the exact lookups fail, verbatim -latest ids keep resolving directly (no behavior change for them).

Test (TDD, Hard Rule #18)

tests/unit/model-capabilities-mistral-vision-sync-4073.test.ts seeds a synced fixture and asserts the verdict comes via the synced path — the discriminator is attachment (synced sets it true/false; the heuristic leaves it null):

• mistral/pixtral-12b-latest → attachment === true, supportsVision === true (RED before: null)
• mistral/ministral-8b-latest → attachment === false, supportsVision === false (synced false wins; RED before: null)
• regression: pixtral-large-latest / bare pixtral-12b still resolve directly
• guard: unknown -latest id never fabricates a match (stays null)

RED 2/4 → GREEN 4/4. Related suites: 89/89 pass (model-capabilities-registry, models-catalog-route, combo-vision-aware-routing, modelsDevSync, vision-detection-consistency #4072). typecheck:core, eslint, check:file-size all clean.

Out of scope (documented follow-up)

The models.dev sync is currently manual-only (Settings → models.dev) with no scheduled refresh, so a fresh instance still relies on the #4071 heuristic until that sync runs. A periodic-refresh cadence is left as a separate follow-up rather than added here.

Follow-up to #4071; sibling of #4072.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[gemini-code-assist]

Code Review

This pull request resolves issue #4073 by ensuring that models.dev synced vision metadata can be correctly resolved for Mistral -latest aliases (e.g., mapping pixtral-12b-latest to pixtral-12b). It introduces a stripLatestAlias helper and updates getSyncedCapabilityForResolved to fallback to stripped aliases when exact matches are not found. Comprehensive unit tests have been added to verify this behavior. The feedback suggests optimizing the candidate loop in getSyncedCapabilityForResolved to prevent redundant lookups and duplicate iterations.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

To avoid redundant database lookups and duplicate iterations, we can optimize the candidate loop:

1. Skip iterating over null values.
2. Avoid duplicate lookups when model === rawModel.
3. Avoid looking up base if it is equal to canonical (which was already looked up and missed on lines 222-225).

  const candidates = [];
  if (model) candidates.push(model);
  if (rawModel && rawModel !== model) candidates.push(rawModel);

  for (const candidate of candidates) {
    const base = stripLatestAlias(candidate);
    if (base && base !== model && base !== rawModel && base !== canonical) {
      const byAlias = getSyncedCapability(provider, base);
      if (byAlias) return byAlias;
    }
  }