docs(zen): add grok-composer-2.5-fast, grok-4.3, grok-4.20-reasoning to model table

[lidge-jun]

Issue for this PR

Closes # N/A — model addition, no tracked issue.

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Adds three xAI/Grok models to the Zen model table in packages/web/src/content/docs/zen.mdx (and all 17 translated variants):

Model	ID	Why
Grok Composer 2.5 Fast	grok-composer-2.5-fast	New coding composition model, live on /v1/chat/completions
Grok 4.3	grok-4.3	xAI's flagship model, was missing from the table
Grok 4.20 Reasoning	grok-4.20-0309-reasoning	Deep reasoning model, was missing from the table

The table previously only had grok-build-0.1. These models are available through the xAI API and users should be able to find them when configuring opencode/<model-id>.

No runtime code changes — transform.ts already handles all Grok models correctly via the if (id.includes("grok")) return {} catch-all on line 671 (no reasoning effort support except grok-3-mini).

I verified grok-composer-2.5-fast is live by making direct API calls with an OAuth session:

• Chat completions: HTTP 200 with reasoning_content in response
• Function calling (tool_use): returns tool_calls with forced tool_choice
• Pricing: ~$1.00/M input, ~$2.00/M output (cheaper than grok-4.3)
• Note: not listed in /v1/models endpoint but responds to direct calls

How did you verify your code works?

1. Confirmed all 18 zen.mdx files were updated with consistent model entries
2. Verified the existing transform.ts handles grok-composer correctly (matches the grok catch-all)
3. Tested the model live via xAI API — chat, tool_use, reasoning all work

Screenshots / recordings

Documentation-only change — no UI changes.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

This model is already adopted by progrok, ima2-gen, hermes-agent PR #36968, and openclaw PR #89190.

[lidge-jun]

Detailed model information

grok-composer-2.5-fast

Grok Build's composer model for agentic code composition.

API probe results (June 2, 2026):

Feature	Status
/v1/chat/completions	✅ HTTP 200
tool_use (function calling)	✅ Returns tool_calls with forced tool_choice
reasoning_content field	✅ Returned in response
System prompt	✅
Multi-turn	✅
Temperature control	✅
stop sequences	❌ 400 (unsupported)
/v1/models listing	❌ Hidden model

Pricing (from cost_in_usd_ticks): ~$1.00/M input, ~$2.00/M output

grok-4.3

xAI's flagship model. $1.25/M input, $2.50/M output, 1M context window. Already available via the xAI API but was missing from the Zen table.

grok-4.20-0309-reasoning

Deep reasoning model with extended thinking. $1.25/M input, $2.50/M output. Supports reasoningEffort — but transform.ts already handles Grok reasoning variants correctly, so no code change needed.

No runtime code changes

transform.ts line 671: if (id.includes("grok")) return {} catches all Grok models that don't match the grok-3-mini specific reasoning path. This PR only adds docs.

Adoption context

Already adopted by progrok, ima2-gen, hermes-agent PR #36968, and openclaw PR #89190.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍