
GPT-5.6 Sol's official price is $4 per million input tokens and $20 per million output tokens — and that figure is itself a promotional price [OI] has said will run at least through November 21, 2026. On the AI compute trading platform, 11 sellers list 12 discount tiers, the deepest at 0.077 of list price: an effective $0.031 for input and $0.154 for output, or 99.23% off. Data as of November 16, 2026.
GPT-5.6 Sol is the flagship of the GPT-5.6 line, aimed at complex reasoning, coding and agent work, with particular strength in command-line and multi-step coding. It opens a 1.05 million-token context window, outputs up to 128K tokens at a time, and supports vision input, tool calls, structured output and context caching — a model built for heavy reasoning.
Official pricing has three tiers: $4 per million input tokens, $20 per million output tokens, and $0.4 per million for cache reads. Output costs five times input, the usual ratio for text models — generating is pricier than understanding. One thing worth flagging: that $4 input price is already a promo. The vendor's docs state it runs "at least through 2026-11-21," so it could revert afterward. Don't treat today's official number as a permanent anchor.
Sellers on the platform set their own prices against the official rate, expressed as a percentage saved. Right now 11 sellers list GPT-5.6 Sol capacity, three of them certified, spread across 12 discount tiers.
The deepest tier carries a buyer_percent_off of 99.23%, meaning you pay official × (100 − 99.23)%, or just 0.77% of list. In real numbers: input $4 × 0.77% = $0.031 per million tokens; output $20 × 0.77% = $0.154 per million tokens.
Put plainly: a million input tokens cost $4 officially and a little over three cents here; a million output tokens cost $20 officially and about 15 cents here. None of this is a time-limited subsidy — it is sellers listing idle capacity at their own prices, which is exactly why the platform can undercut direct official access.
The 12 tiers the 11 sellers have posted, deepest first: 99.23%, 98.35%, 97.8%, 97.58%, 95.05%, 94.5%, 89%, 78%, 72.5%, 67%, 56%, 50%. The chart below plots them by depth — a taller bar means a deeper discount and a lower effective price.
One key point: a deeper discount does not mean a better seller. Each tier is a price one seller posted, and different sellers may list different discounts for the same model. The deepest 0.077 listing is not necessarily more reliable than a 0.5 one — your order is priced by the effective rate, but success rate and stability depend on the seller's reputation and history. That is why the platform flags certified sellers separately. Price is only one dimension of choosing a source; don't fixate on the lowest number.
GPT-5.6 Sol's promotional price comes with two extra billing rules you should read before ordering:
First, long inputs cost more. Input past 272K tokens is billed at 2× the input rate, and output at 1.5×. The context window opens to 1.05 million, but once you actually fill it with a long document, the extra portion gets more expensive — long context is the selling point, and it also costs more.
Second, cache writes cost extra. Cache writes bill at 1.25× the uncached input rate, or $5 per million. Cache reads are cheap ($0.4 per million), but writing into the cache is a paid step in itself — caching only pays off when you repeatedly hit the same context.
GPT-5.6 Sol suits three kinds of work: complex reasoning, multi-step coding, and long-context agent tasks. A 1.05M context window plus a 128K output cap make it a fit for long documents and large codebases.
Three honest caveats. First, the platform still has very few calls on this model, so the health panel has no stable latency or success-rate figures yet — validate with a small traffic slice before production. Second, the promo may end November 21, so lock in cost sooner rather than later. Third, as noted, deepest discount and seller reliability are two different things.
Three things you can act on today: search GPT-5.6 Sol in the model plaza for live tiers and seller counts; check its model docs for the billing rules (long-input multiplier, cache writes) and runnable examples; then trial it on a small traffic slice before moving production over.
Every price in this article is a snapshot from November 16, 2026. Sellers post and withdraw asks constantly, so go by the live numbers in the price table and plaza at order time.
GPT-6.2 cuts list prices 20%, yet platform discounts run deeper: gpt-6-astra trades at under 1% of list. What this week's AI price war means for buyer
Prompt caching stores a repeated prompt prefix and bills cache reads at 5-12% of normal input. Real prices, hit conditions, and how to cut your API bi
Twelve Claude versions are on sale. The newest opus-5-5 costs $8.91 a month; the oldest opus-4-6 costs $49.50. Full table and a pick for each workload
gpt-5.6-terra and deepseek-v4.1-flash win on price and reliability; gemini-3.8-flash is fastest. Effective prices, measured speed, and a three-tier pi
Point your base URL at the platform and one key covers all 138 on-sale models across 51 vendors, in both OpenAI and Anthropic request formats. Copy-pa