Update GPT-5.6 Sol pricing for OpenAI's Aug 21 promo cut (openai + openrouter)

[Nauxie]

What

OpenAI cut GPT-5.6 Sol developer pricing by 20% on input and ~33% on output, effective Aug 21, 2026 (promotional, running at least through Nov 21, 2026). Reuters coverage here.

pricing/openai.json — gpt-5.6-sol + gpt-5.6-sol-2026-07-09

Per 1M tokens, from OpenAI's pricing page:

Tier	Input	Cached input	Output
Standard (≤272k)	$5 → $4	$0.50 → $0.40	$30 → $20
Long context (>272k)	$10 → $8	$1 → $0.80	$45 → $30
Batch / Flex (≤272k)	$2.50 → $2	$0.25 → $0.20	$15 → $10
Priority (≤272k)	$10 → $8	$1 → $0.80	$60 → $40

All tiers (including the data-residency region block) scale uniformly: input-side ×0.8, output ×⅔, which matches the published table exactly. Cache-write keeps its 1.25× input ratio. additional_units (web_search, file_search) unchanged.

pricing/openrouter.json — openai/gpt-5.6-sol, -pro, and :batch variants

Synced to OpenRouter's live /api/v1/models response: $2/M input, $10/M output (batch $1/$5). OpenRouter layers its ongoing 50%-off non-BYOK promo on top of OpenAI's new base price; this file already tracked the promo-discounted listed price, so these values mirror the current API response verbatim.

Deliberately not changed

• azure-openai / azure-ai — no evidence Microsoft has matched the cut yet (trackers and the Azure retail Prices API still show old rates; the promo is scoped to OpenAI's API per the announcement). Should follow up if/when Azure updates.
• bedrock / bedrock-mantle — AWS still lists Sol at $5/$30 (+10% regional uplift); AWS sets Bedrock pricing independently.
• lightning-ai — no published change found.

Validation

• python3 scripts/check_duplicate_keys.py clean
• prettier --check and eslint --quiet pass on both files
• Diff touches only "price" lines (128 in openai.json, 14 in openrouter.json)

🤖 Generated with Claude Code