GPT-5.6 Sol’s limited-time price cut vs Chinese models
A dated public-price snapshot for the 2026-09-01 to 2026-11-30 window. This is a rate comparison, not a quality benchmark or a claim that every model is available through AIWave.
The short answer
Even with the Sol promotion, it is about 2.09× the input rate and 3.48× the output rate of deepseek-v4-pro. Against deepseek-v4-flash, the gaps are about 6.27× input and 10.45× output. So the “2–3.5×” conclusion is accurate only for the v4-pro comparison; flash is a wider price gap.
Price table
| Model | Input / 1M tokens | Output / 1M tokens | Public source and date |
|---|---|---|---|
| GPT-5.6 Sol | $4.00 | $20.00 | OpenAI model page; promotion update 2026-08-21, checked 2026-09-25 |
| deepseek-v4-pro | $1.914 | $5.742 | AIWave pricing; effective 2026-08-27, checked 2026-09-25 |
| deepseek-v4-flash | $0.638 | $1.914 | AIWave pricing; effective 2026-08-27, checked 2026-09-25 |
All figures are USD per million tokens. Cached-input rates, long-context tiers, billing-group multipliers, taxes, and retries are excluded from this simple comparison.
What this means for an agent builder
- Use the live catalog and your effective billing group before estimating a workflow.
- Reserve the higher-priced model for tasks where its observed quality or tool reliability justifies the difference.
- For routine summaries and routing experiments, test a smaller model with the same prompt and evaluation set.
Start with the routing recipe, then check request recovery for operational failure handling.
Sources and method
Sol: OpenAI’s official model page and its 2026-08-21 promotion update. AIWave: the public pricing JSON and pricing page, both checked 2026-09-25. Multiples are simple division of listed input and output rates, rounded to two decimals.