Update Studio Code agent's supported Sonnet model to Sonnet 5

[youknowriad]

Related issues

• Related to #

How AI was used in this PR

Claude Code located the central model registry, applied the model swap, updated the affected unit tests, and drafted this PR. The author reviewed the diff.

Proposed Changes

The Studio Code agent's selectable Anthropic Sonnet option is bumped from Sonnet 4.6 to Sonnet 5, and Sonnet 5 becomes the default model. Users picking Sonnet in the model picker (and new sessions that don't pin a model) now get the newer model. Existing sessions that recorded the old claude-sonnet-4-6 id gracefully fall back to the default via resolveSessionModel, so nothing breaks for in-flight chats.

All model-driven UI (the /model slash command, both composer model pickers, the AiModelId type) derives from the single AI_MODELS array, so this is a one-line registry change plus test updates.

Testing Instructions

• Open the Studio Code agent, open the model picker, and confirm Sonnet 5 appears and is selectable.
• Run a turn against Sonnet 5 and confirm the request succeeds against the configured provider.
• npm test -- apps/cli/ai/tests/slash-commands.test.ts apps/cli/ai/tests/pi-runtime.test.ts

Note: unit tests, lint, and typecheck could not be run green in the authoring environment due to a pre-existing local toolchain issue (Node 20 vs required ≥22, unbundled @studio/common/Playground deps, a @wordpress/ui version mismatch) unrelated to this change. Please rely on CI.

Pre-merge Checklist

• Have you checked for TypeScript, React or other console errors?
• Confirm the claude-sonnet-5 model id matches what the provider expects.

[wpmobilebot]

📊 Performance Test Results

Comparing b73ed7d vs trunk

app-size

Metric	trunk	b73ed7d	Diff	Change
App Size (Mac)	1316.77 MB	1316.77 MB	0.00 MB	⚪ 0.0%

site-editor

Metric	trunk	b73ed7d	Diff	Change
load	1112 ms	1088 ms	24 ms	⚪ 0.0%

site-startup

Metric	trunk	b73ed7d	Diff	Change
siteCreation	6517 ms	6494 ms	23 ms	⚪ 0.0%
siteStartup	1863 ms	1859 ms	4 ms	⚪ 0.0%

---

Results are median values from multiple test runs.

Legend: 🟢 Improvement (faster) | 🔴 Regression (slower) | ⚪ No change (<50ms diff)

[youknowriad]

I built a couple of sites locally, quality seems good.
I run the evals too.

[youknowriad]

I've been running the evals for hours now. My tests suggest that Sonnet 4.6 is actually better than Sonnet 5.0. It passes our evals more consistently. Can anyone confirm? cc @chubes4 @sejas

I'm starting to doubt whether we should make this update.

[youknowriad]

More specifically, the slideshow and shop evals (so the plugin recommendation related evals)

[chubes4]

@youknowriad I will run an eval soon, but anecdotally I did see some posts from users on X about how Sonnet 5 ends up more expensive than Opus on long-running coding tasks and is better suited for knowledge work than coding.

[youknowriad]

the funny thing is that I'm not even comparing it with Opus but to Sonnet 4.6 here.

[youknowriad]

I run more evals, Sonnet 5 is definitely flakier than Sonnet 4.6, especially for the "slideshow" eval.

[chubes4]

I ran a couple of evals via studio-eval with Sonnet 5 vs Sonnet 4.6.

Sonnet 5 timed out, but 4.6 completed successfully on both runs.