OpenAI · GPT-5.6
GPT-5.6 Sol
OpenAI’s main reasoning model for complex knowledge work, research, coding and tasks that need more deliberate analysis.
Treat the effort control as the product. Run one real job at two levels and keep the cheapest level that survives review.
Use it for
- Conflicting evidence and multi-step analysis
- Complex coding or computer-use work
- Difficult research and design decisions
Skip the extra latency for simple edits, obvious lookups and work where the source, not deeper reasoning, is missing.
Effort without the jargon
A considered answer should be enough and you want a sensible default for non-trivial work.
The task has competing constraints, several steps or a costly error that justifies more thought.
A Pro plan user has a genuinely difficult task and will compare the gain against time and allowance.
Advanced recordCost, limits, privacy and operations
Use ChatGPT effort controls for difficult work, but budget the API separately: Sol has a published million token context, a long context multiplier and tool charges that can change the total task cost.
- API cost
- Officially disclosedOpenAI API price: US$4 per million input tokens, US$0.40 cached input and US$20 per million output, a promotional rate the model and pricing pages hold at least through 21 November 2026 (the earlier US$5/US$30 standard rate is no longer shown on the Sol page and now describes GPT-5.5). Requests above 272,000 input tokens are priced at twice the input rate and one and a half times the output rate for the full request; tool charges can be separate.
- Context window
- Officially disclosed1,050,000 tokens
- Maximum output
- Officially disclosed128,000 tokens
- Knowledge cutoff
- Officially disclosed16 February 2026
- Recorded inputs
- Text · Images
- Recorded output
- Text
Data handling is a product decision
Depends on surfaceOpenAI says API data is not used for training unless the organisation opts in; retention varies by endpoint and controls. ChatGPT has separate product and workspace settings.
Operational constraints
- ChatGPT availability, reasoning allowances and fallbacks still depend on plan and workspace settings.
- API requests above 272,000 input tokens use the published long context multiplier for the whole request.
- The gpt-5.6 API alias routes to Sol, while ChatGPT model and effort selection are product-managed and can differ from a pinned API request.
- Web search, file search, computer use and other tools can add separate charges and permission requirements.
Unknown means the cited official record does not disclose a safe value. It is not an estimate. Prices are provider list prices in USD where stated and can change before this record's review date.
Keep the evidence separate
OpenAI says Sol powers Medium, High and Extra High in ChatGPT and targets complex work across coding, research, science, computer use and design.
His early use found Sol clearly competent but not obviously better than Claude Fable for his complex coding work; he also warns that price per token misses differences in reasoning volume.
Read the independent source ↗No Human Bit result is claimed; use this as evaluation guidance only.
Not scheduled
No Human Bit result is claimed for this model yet.