feat(models): add Opus 4.8 and Sonnet 5

[dimakis]

Summary

• Add Claude Opus 4.8 + Max variant to available models
• Add Claude Sonnet 5
• Remove Opus 4.7 and Opus 4.5 (not provisioned on new Vertex project itpc-ca-638a2a9c7c)
• Update model-spec tests

Test plan

• model-spec.test.ts passes (11/11)
• Verify model selector renders correctly on mobile and desktop
• Verify Opus 4.8 sessions work end-to-end

🤖 Generated with Claude Code

[dimakis]

Centaur Review

Found 3 issue(s) (2 warning).

server/chat.ts

Model list updates are consistent across all three picker locations (server, mobile, desktop). Main issue: the parseModelSpec tests and a JSDoc comment still reference the removed claude-opus-4-7 model ID.

• 🟡 regressions (L378): JSDoc comment still references claude-opus-4-7:max — should be updated to claude-opus-4-8:max to match the new model list. [fixable]

server/__tests__/model-spec.test.ts

Model list updates are consistent across all three picker locations (server, mobile, desktop). Main issue: the parseModelSpec tests and a JSDoc comment still reference the removed claude-opus-4-7 model ID.

• 🟡 regressions (L10): All four parseModelSpec test cases (lines 10, 17, 24, 28) still use claude-opus-4-7 as test input. The function is a generic string splitter so they pass, but they reference a model that no longer exists in AVAILABLE_MODELS. Should be updated to claude-opus-4-8 for consistency — the resolveThinking tests in the same file were updated but these were missed. [fixable]

frontend/src/lib/constants.ts

Model list updates are consistent across all three picker locations (server, mobile, desktop). Main issue: the parseModelSpec tests and a JSDoc comment still reference the removed claude-opus-4-7 model ID.

• 🔵 regressions (L12): The multi-line comment explaining the DEFAULT_MODEL choice (lines 12-16) references "Opus 4.7" four times. Now that the picker offers 4.8, these references are stale and should say "4.8". [fixable]

[dimakis]

Centaur Review

Found 3 issue(s) (1 warning).

server/__tests__/ws-handler-v2.test.ts

Straightforward model roster update. The pickers and AVAILABLE_MODELS are consistent, but 6 stale claude-opus-4-7 references remain in test fixtures across two test files.

• 🟡 regressions: 4 stale references to claude-opus-4-7 remain at lines 2367, 2376, 2438, 2447. While these tests still pass (the model string is opaque passthrough data), they should be updated to claude-opus-4-8 for consistency with the rest of the PR. [fixable]

packages/client/__tests__/store.test.ts

Straightforward model roster update. The pickers and AVAILABLE_MODELS are consistent, but 6 stale claude-opus-4-7 references remain in test fixtures across two test files.

• 🔵 regressions: 2 stale references to claude-opus-4-7 remain at lines 795 and 799. Same situation — opaque test fixture value that still passes, but should be updated for consistency. [fixable]

server/__tests__/model-spec.test.ts

Straightforward model roster update. The pickers and AVAILABLE_MODELS are consistent, but 6 stale claude-opus-4-7 references remain in test fixtures across two test files.

• 🔵 missing_tests (L55): resolveThinking has no test case for claude-sonnet-5. The existing claude-sonnet-4-6 test covers the .includes('sonnet') branch, but since Sonnet 5 is a newly-added model with a different naming pattern (no minor version), an explicit test would guard against future regressions if the matching logic changes. [fixable]

[dimakis]

Centaur Review

LGTM — no issues found.