[BOT ISSUE] Cost update: claude-sonnet-4-20250514 context window wrong (1M → 200K)

[github-actions]

Summary

The claude-sonnet-4-20250514 entry in model_list.json has an incorrect context window. It is set to 1,000,000 tokens but should be 200,000 tokens per the Anthropic documentation.

The same issue applies to the aliased entry claude-4-sonnet-20250514 if present.

Current vs correct values

Field	Current Value	Correct Value
max_input_tokens	1000000	200000

Verification

• Source 1: Anthropic models page (https://docs.anthropic.com/en/docs/about-claude/models) — the legacy models table shows Claude Sonnet 4 (deprecated) with a 200K token context window (~150k words, ~680k unicode characters)
• Source 2: The same page shows that only Claude Opus 4.6+ and Claude Sonnet 4.6 have 1M context windows. All earlier models (Sonnet 4.5, Opus 4.5, Opus 4.1, Sonnet 4, Opus 4) have 200K.
• ID format: claude-sonnet-4-20250514 is the correct dated snapshot ID per Anthropic docs
• Recent commits: No recent commits correcting this value

Additional context

This model is deprecated (retirement date: June 15, 2026), but the context window should still be correct while it remains in the catalog. The catalog already correctly has deprecation_date set.

[github-actions]

Prepared catalog updates for claude-sonnet-4-20250514, claude-4-sonnet-20250514. Updated models: claude-sonnet-4-20250514, claude-4-sonnet-20250514.

Verification sources: https://platform.claude.com/docs/en/docs/about-claude/models
Verified metadata applied:

• claude-sonnet-4-20250514: format=anthropic, flavor=chat, displayName=Claude 4 Sonnet, max_input_tokens=200000, max_output_tokens=64000, pricing(input/output per 1M)=3/15, input_cache_read_cost_per_mil_tokens=0.3, input_cache_write_cost_per_mil_tokens=3.75, providers=anthropic, reasoning=true, reasoning_budget=true, multimodal=true, deprecation_date=2026-05-14
• claude-4-sonnet-20250514: format=anthropic, flavor=chat, displayName=Claude 4 Sonnet (old naming format), max_input_tokens=200000, max_output_tokens=64000, pricing(input/output per 1M)=3/15, input_cache_read_cost_per_mil_tokens=0.3, input_cache_write_cost_per_mil_tokens=3.75, providers=anthropic, multimodal=true, deprecated=true

Opened #691 to apply the catalog update.