feat: add gitlab duo-chat-opus-5 model

[vglafirov]

What

Adds providers/gitlab/models/duo-chat-opus-5.toml for the GitLab provider's new duo-chat-opus-5 model id (Claude Opus 5) so it shows up in opencode's model picker.

Details

Uses base_model = "anthropic/claude-opus-5" to inherit provider-agnostic metadata (context 1,000,000 / output 128,000 tokens, modalities text, image, pdf input / text output, temperature = false, reasoning options), matching the shape already used for duo-chat-sonnet-5 and duo-chat-fable-5. Only overrides:

• name — GitLab's "Agentic Chat (...)" branding
• reasoning_options — trimmed to the effort variant (no toggle), matching Opus 5's base reasoning options
• [cost] — zeroed out (billed upstream via the GitLab AI Gateway proxy)

Source: ai_gateway/model_selection/models.yml in gitlab-org/modelops/applied-ml/code-suggestions/ai-assist — claude_opus_5 entry (proxy_provider: anthropic, params.model: claude-opus-5).

Companion change: vglafirov/gitlab-ai-provider!33 adds the duo-chat-opus-5 / duo-workflow-opus-5 mappings in the provider itself.

Verification

bun validate passes and confirms duo-chat-opus-5 is present in the generated model registry, correctly inheriting all fields from anthropic/claude-opus-5.

[github-actions]

Action items

• [high] [violation] providers/gitlab/models/duo-chat-opus-5.toml:1 - Check: New models must use base_model when matching models/ metadata exists. Why: models/anthropic/claude-opus-5.toml already defines the provider-agnostic facts for Claude Opus 5, and other GitLab entries (e.g. duo-chat-sonnet-5) correctly inherit via base_model. This file instead inlines those facts and drifts from the canonical metadata (knowledge = "2026-01-31" vs "2026-05"; release_date/last_updated "2026-07-27" vs "2026-07-24"). Action: Rewrite to base_model = "anthropic/claude-opus-5" and keep only provider-specific fields/overrides (name, cost, reasoning_options, and any intentional limit overrides).

[github-actions]

Action items

• [medium] [possible mistake] providers/gitlab/models/duo-chat-opus-5.toml:3 - Check: reasoning_options must reflect controls exposed on this provider’s request surface (npm = "gitlab-ai-provider"), not upstream Anthropic alone. Why: The entry sets effort values low/medium/high/xhigh/max, matching native Anthropic claude-opus-5, but providers/gitlab/provider.toml documents that gitlab-ai-provider builds fixed request bodies with no reasoning passthrough, so those raw Anthropic shapes are not available through this integration. Same-surface GitLab Opus peers (duo-chat-opus-4-5 through duo-chat-opus-4-8) all use reasoning_options = []. The PR body only cites model-id mapping / models.yml presence, not GitLab effort control. Action: Use reasoning_options = [] unless you can show this model now exposes caller effort via the GitLab path (package change, raw request field, or docs). If effort is newly supported, cite that evidence in the PR body and align provider.toml’s reasoning note.