AI Tools
GPT-5.6 Sol price cut: what does the model cost now?
OpenAI cut GPT-5.6 Sol price to $4 per million input tokens and $20 per million output tokens, down from $5 and $30, with the new rates marked as promotional pricing at least through November 21, 2026. GPT-5.6 Sol is OpenAI's flagship frontier model; the cut also lowers cached-input pricing to 40 cents per million tokens.
What matters
- OpenAI cut GPT-5.6 Sol's direct API rate to $4 per million input tokens and $20 per million output tokens, down from $5 and $30, verified on the official pricing page August 25, 2026.
- The promotional pricing is guaranteed at least through November 21, 2026; "at least" means no price hike before that date, but OpenAI can reset rates after it.
- Long-context standard requests cost double: $8 input, $0.80 cached input, $10 cache writes, $30 output per million tokens.
- Batch API runs GPT-5.6 Sol at $2 input and $10 output per million tokens on short context, half the standard tier.
- OpenRouter still undercuts the new direct rate at $2.50/$15 per million tokens (verified August 19, 2026), while DeepSeek V4 Pro output costs $1.98 off-peak and $3.96 peak.
What is GPT-5.6 Sol?
GPT-5.6 Sol is OpenAI's flagship frontier model and the top tier of the GPT-5.6 family, which also includes GPT-5.6 Terra and GPT-5.6 Luna. OpenAI's pricing page lists gpt-5.6-sol as the most expensive model in the current lineup: $4 per million input tokens and $20 per million output tokens after the August 2026 reduction, versus $2/$12 and $0.20/$1.20 for Terra and Luna.
It is the model that resellers such as OpenRouter carry under the gpt-5.6-sol name, and the one API buyers benchmark when they ask whether frontier-level quality justifies the bill. The price cut this week changes that calculation for every team that priced the model a month ago.
GPT-5.6 Sol price cut: what changed
As of August 25, 2026, the OpenAI pricing page shows gpt-5.6-sol at $4.00 per million input tokens, $0.40 for cached input, $5.00 for cache writes, and $20.00 for output in the standard short-context tier. Before the cut, verified on August 19, 2026, the direct rate was $5 input and $30 output per million tokens, the baseline OpenRouter used when it priced the model at exactly half.
- Standard short context: $4 input, $0.40 cached input, $5 cache writes, $20 output.
- Standard long context: $8 input, $0.80 cached input, $10 cache writes, $30 output.
- Batch short context: $2 input, $0.20 cached input, $2.50 cache writes, $10 output.
The reduction is 20% on input and 33% on output. Long-context and batch tiers dropped by the same proportions, so the relative cost of context length and batch discount is unchanged. For comparison, the August 19 companion analysis on this site covers how the OpenRouter resale rate related to the old direct price.
What the November 21 deadline means
OpenAI labels the new rates as promotional pricing and states it is available at least through November 21, 2026. The phrase "at least" is the operative detail: OpenAI guarantees the price will not rise before that date, but it can either extend the promo or reset rates after it.
For a buyer planning a multi-month project, the date matters. A workload that starts at the promo rate could see its input or output price change mid-project. Resellers such as OpenRouter set their own margins on top of the OpenAI rate, so a direct cut does not force a matching resale cut; it simply narrows the reseller discount.
GPT-5.6 Sol vs OpenRouter and DeepSeek
Direct pricing is now closer to resale pricing, but the spread remains. OpenRouter listed gpt-5.6-sol at $2.50 per million input tokens and $15 per million output tokens on August 19, 2026, exactly half of the old OpenAI rate. Against the new $4/$20 direct price, OpenRouter is 37.5% cheaper on input and 25% cheaper on output. Buyers who route through OpenRouter still pay less per token, in exchange for depending on a reseller's availability and rate stability.
DeepSeek undercuts both. DeepSeek V4 Pro bills $0.66 per million input tokens (cache miss) and $1.98 per million output tokens off-peak, per the DeepSeek pricing page on August 25, 2026. At peak hours those rise to $1.32 and $3.96, still about one fifth of GPT-5.6 Sol's new output rate. DeepSeek is not a drop-in equal in quality, but for cost-sensitive inference it remains the price anchor of this segment.
The decision frame: pay OpenAI direct for first-party access and the full toolchain, route through OpenRouter for the same model at a lower token price, or move to DeepSeek-class models when the workload tolerates a different frontier and the bill is the deciding factor.
Who should buy direct now, and who should wait
Teams already running on the OpenAI API should treat the cut as a free discount: no code change, no migration, just a smaller line item on the invoice. Workloads with heavy prompt caching benefit from the cached-input rate dropping to $0.40 per million tokens.
Teams choosing a model this week should price the November 21 expiry into the comparison. A project that plans to run past that date carries price-change risk that DeepSeek-class flat pricing does not. Anyone comparing raw token cost should also keep the tier straight: $4/$20 is the short-context standard rate, and long-context requests double to $8/$30.
One honest catch: the promo applies to GPT-5.6 Sol only. Terra and Luna, the cheaper members of the family, are not marked as promotional on the same page, so the family's internal price hierarchy is unchanged.
At a glance
| Tier | Input | Cached input | Cache writes | Output |
|---|---|---|---|---|
| Standard, short context (new) | $4.00 | $0.40 | $5.00 | $20.00 |
| Standard, long context (new) | $8.00 | $0.80 | $10.00 | $30.00 |
| Batch, short context (new) | $2.00 | $0.20 | $2.50 | $10.00 |
| Direct rate before cut (Aug 19) | $5.00 | $0.50 | $6.25 | $30.00 |
| OpenRouter resale (Aug 19) | $2.50 | $0.25 | $3.125 | $15.00 |
| DeepSeek V4 Pro, off-peak (Aug 25) | $0.66 | n/a | n/a | $1.98 |
FAQ
When does the GPT-5.6 Sol price cut expire?
OpenAI states the promotional pricing is available at least through November 21, 2026. The phrase "at least" means the rate cannot rise before that date; after it, OpenAI can extend the promo or set a new price.
Is GPT-5.6 Sol cheaper than DeepSeek?
No. DeepSeek V4 Pro bills $1.98 per million output tokens off-peak and $3.96 peak, versus $20 for GPT-5.6 Sol short-context output. GPT-5.6 Sol is a different class of frontier model; the cut narrows the gap but does not close it.
What is GPT-5.6 Sol?
GPT-5.6 Sol is OpenAI's flagship frontier model and the most expensive member of the GPT-5.6 family, which also includes GPT-5.6 Terra and GPT-5.6 Luna. After the August 2026 cut it costs $4 per million input tokens and $20 per million output tokens on the standard tier.
Related reading
GPT-5.6 Sol pricing: OpenRouter vs OpenAI, OpenRouter alternatives, DeepSeek API pricing, GLM-5.3 pricing