OpenAI guarantees promotional GPT-5.6 Sol API pricing through at least November 21

Promotional rates locked in

OpenAI is holding GPT-5.6 Sol at promotional pricing through November 21, 2026 at minimum, according to the company's API pricing documentation, which drew attention after reaching the Hacker News front page on August 24. For developers building on the model, the commitment translates directly into per-token API costs they can budget around for the coming months.

Under the short-context rates listed on the page, GPT-5.6 Sol bills at $4.00 per million input tokens, $0.40 per million cached input tokens, $5.00 per million cache-write tokens, and $20.00 per million output tokens. Long-context requests are charged at a separate, higher tier: $8.00 input, $0.80 cached input, $10.00 cache writes, and $30.00 output per million tokens.

OpenAI labels these numbers as promotional but does not publish the standard rates they are discounted from, so the exact size of the reduction cannot be verified from the documentation alone. What is explicit is the wording around the deadline: the pricing holds "at least through" late November, which leaves room for an extension without promising one.

Where Sol sits in the GPT-5.6 lineup

The same page positions Sol in the middle of the GPT-5.6 family. Terra lists at $2.00 input and $12.00 output per million tokens for short context, doubling to $4.00 and $18.00 for long context. Luna, the cheapest of the three, comes in at $0.20 input and $1.20 output per million tokens for short context. At the top end, gpt-5.6-cyber is priced at $12.50 input and $75.00 output per million tokens, with no long-context tier listed.

Cached input runs at exactly one tenth of standard input pricing across the family, which makes prompt caching a meaningful lever at Sol's rate card, particularly for long-context workloads where input costs double.

OpenAI also exposes Sol through its Daybreak alias system. According to the documentation, daybreak-blue-latest currently resolves to gpt-5.6-sol, while daybreak-red-latest resolves to gpt-5.6-cyber. The company notes that these aliases will be repointed to newer models as the Daybreak program releases them, with pricing following whichever model sits underneath. That is worth noting for anyone pinning costs to an alias rather than a pinned model ID, since a silent repoint could change both capability and spend.

Other changes on the pricing page

Several additional items on the page affect API bills and workflows:

Data residency: regional processing endpoints carry a 10% uplift for models released on or after March 5, 2026 that are eligible for data residency.

Fast mode: priority processing was renamed to Fast mode on July 30, 2026. Both service_tier values, "priority" and "fast", are accepted in API requests, so existing integrations should keep working through the transition.

Fine-tuning wind-down: OpenAI says it is shutting down its fine-tuning platform. New users can no longer access it, and existing users can create training jobs only for the coming months. Fine-tuned models will remain available for inference until their base models are deprecated. The pricing page points to a separate document for the full timeline but does not spell out dates there.

Bedrock billing: OpenAI models available through Amazon Bedrock are billed by AWS and may differ from OpenAI's direct pricing.

The page also lists rates for other product lines, including the gpt-realtime-2.1 audio and text models, gpt-image-2, the sora-2 video generation tiers, and a set of transcription models, none of which are covered by the Sol promotion.

Why it matters

Per-token pricing is a core planning input for anyone shipping LLM features, and a guaranteed rate through late November removes one variable from Q3 and Q4 budgets. The open-ended expiry language means teams should still plan for a possible step up after November 21, since OpenAI has committed to nothing beyond that date.

Two second-order effects deserve attention. First, anyone using the Daybreak aliases rather than explicit model IDs is exposed to cost changes whenever OpenAI repoints them, regardless of the Sol promotion. Second, the fine-tuning platform wind-down gives teams relying on custom-trained models a narrowing window to migrate, with the documentation offering only a vague timeline of "coming months" for existing users.