OpenAI cuts GPT-5.6 Sol API and credit pricing by over 20% for three months

OpenAI quietly appended an update to its GPT-5.6 launch page on August 21: Sol, the flagship model in the GPT-5.6 family, now has a temporary price cut on API and credit billing. The move is the third price adjustment in the family in two months — and it matters for anyone budgeting GPT-5.6 Sol usage over Q4 2026.

This is an update note, not a re-coverage of the launch.

What happened

The GPT-5.6 launch page now opens with this line:

“OpenAI dropped the API and credit pricing of GPT‑5.6 Sol by over 20% for the next 3 months.”

The body of the page has not been updated. The “Availability and pricing” section still lists the original base price: Sol is $5 input / $30 output per 1M tokens. No new per-token figure has been published.

This follows two earlier moves in the same family:

• Jul 30, 2026: Luna cut by 80% (to $0.20/$1.20 per 1M tokens), Terra cut by 20% (to $2/$12 per 1M tokens) — per the price-performance post.
• Aug 21, 2026: Sol cut by over 20% for 3 months — per the launch page update.

Why it matters

• Q4 cost planning. API users paying for GPT-5.6 Sol today should expect a lower invoice for approximately three months from Aug 21. Agents, long-running workflows, and batch jobs benefit most from per-token price drops.
• Third price move in two months. The Sol family has gone from launch pricing to three distinct price adjustments in six weeks. That pace signals competitive pricing pressure and OpenAI’s willingness to pass efficiency gains through quickly.

Practical implications

Who benefits: API users running GPT-5.6 Sol for long agent loops, batch inference, and token-heavy workflows. The effective cost per million output tokens drops by at least 20% during the window.

What does NOT change:

• Fast mode pricing remains at 2× the per-token rate for ~2.5× speed on Sol (introduced Jul 30).
• Cache writes are still billed at 1.25× the uncached input rate.
• Cache reads still receive the 90% cached-input discount.
• ZDR behavior and the 30-minute minimum cache life are unchanged.

What to verify before locking in cost forecasts:

• The new Sol per-token number on the platform pricing page.
• Whether “credits” means API billing credits, ChatGPT Work credits, or both.
• The exact end date of the 3-month window.

Risks and caveats

• The exact new per-token price has not been published. The claim is “over 20% for 3 months.” A 20% reduction on the listed $5/$30 would be $4/$24 — but that is arithmetic illustration, not a published figure. Do not treat it as the official new price.
• “Credits” is undefined. The update says “API and credit pricing.” It is not yet clear whether “credits” refers to API billing credits, ChatGPT Work credits, or both.
• The 3-month end date is not published. “Approximately three months from Aug 21, 2026” is the best available framing. The exact expiry date has not been announced.
• The launch page body has not been updated. The “Availability and pricing” section still shows $5/$30. The update note was appended at the top; the page body may or may not be updated later.

What to watch

• The OpenAI platform pricing page for an updated Sol per-token number and an explicit end date for the 3-month window.
• Any clarification on what “credits” means in the update line.
• A possible follow-up post on the price-performance frontier page with updated Sol numbers.

Sources