GPT-5.6 Sol
OpenAI's frontier workhorse for reasoning, coding, and agentic runs.
- Maker
- OpenAI
- Context
- 1.05M tokens
- Pricing
- $5 / 1M tokens / $30 / 1M tokens
Watch the long-context cliff: requests over 272K input tokens are billed at $10 in / $45 out. Same model, double the input rate.
Best for
- Multi-step analysis, trade-off reasoning, and planning
- Long-horizon agentic work and complex coding
- Cross-checking a conclusion you reached with another frontier model
Prompting quirks
Don't tell it to "think step by step"
Reasoning is built in, and there's a reasoning-effort control for depth. Step-by-step boilerplate adds tokens and makes output more verbose, not better. State the goal and the constraints instead.
Stay under 272K input tokens unless you mean it
The window runs to 1.05M, but crossing 272K input tokens moves the whole request to a higher rate. Summarise or chunk first; treat the top of the window as an emergency exit, not a habit.
Give it the success criteria, then get out of the way
"Optimise for X, here are the constraints, here's what done looks like" beats a long procedural script. Let it find the path.
Copy-paste example
Tuned for GPT-5.6 Sol. Swap in your own content where marked.
Here is our Q3 roadmap and the reasoning behind the ranking:
{paste roadmap + rationale}
Find the weakest link in this prioritisation. What are we assuming that,
if false, would flip the order? Give me the single highest-leverage
thing to validate first and why.Facts verified against the official OpenAI docs on 19 August 2026.