GPT-5.6 Sol Price Cut: $4/$20 per Million for 3 Months

OpenAI just made its frontier model dramatically cheaper to run: GPT-5.6 Sol’s API and credit pricing drops by over 20% for the next three months, with output tokens falling a full 33% — from $30 to $20 per million tokens.

Key facts

• New Sol rates (per million tokens): $4.00 input (was $5.00, −20%), $0.40 cached input (was $0.50), $20.00 output (was $30.00, −33%).
• Duration: promotional pricing is available at least through November 21, 2026 — the “next 3 months” per OpenAI’s announcement.
• Where it applies: the API now; rolling out across eligible ChatGPT Work and Codex credit plans.
• Where it doesn’t: Pro, Plus, and Business subscription usage remains unchanged.
• Other tiers on Sol: Batch runs $2/$10; Fast mode runs $8/$40.

What changed

The announcement landed August 21, and OpenAI’s pricing page confirms the new rates plus two details worth knowing before you re-budget:

1. Long prompts cost more: requests above 272K input tokens are billed at 2x input and 1.5x output for the full request — relevant if you’re stuffing near-max context.
2. Cache writes bill at 1.25x the uncached input rate, so cache-heavy agent loops should be modeled accordingly.

This is the third GPT-5.6 repricing in under a month. On July 30, OpenAI cut Terra by 20% ($2.50→$2.00 in, $15→$12 out) and Luna by 80% ($1.00→$0.20 in, $6→$1.20 out), crediting production GPU kernel improvements that lowered serving costs by 20% and speculative-decoding gains of over 15% in token-generation efficiency. Sol was the last tier holding its launch price — until now.

Why it matters

The competitive math moved first. DeepSeek’s V4-Pro shift to peak/off-peak pricing (our coverage) pushed peak-hour frontier rates into play the same week, Google shipped Gemini 3.7 Flash at half of 3.6 Flash pricing, and our API price tracker shows open-weight models closing the gap (comparison here). A 33% output-token cut on the flagship keeps Sol’s quality-per-dollar defensible against both.

For builders, three practical notes: AWS Bedrock followed within a day with matching Sol reductions; the earlier Ultrafast preview remains a separate speed product, not part of this cut; and treat the window as genuinely temporary — OpenAI’s own community thread assumes standard pricing returns after November 21 unless extended.

What to watch

Whether the promo becomes permanent when Terra and Luna cuts get evaluated at renewal — OpenAI has historically converted efficiency gains into standing prices. And whether Anthropic responds before Claude Sonnet 5’s intro pricing ends August 31.

Official source

OpenAI Developers — GPT-5.6 Sol pricing; announcement thread (August 21, 2026).