GPT-6.1 Sol Has No none or minimal Reasoning, and Its Cache Writes Cost More Than Plain Input

GPT-6.1 Sol launched September 29 at one fifth of Astra's price, but three details in its documentation change how you should call it: reasoning.effort starts at low and defaults to medium, cache writes are billed at 1.25 times the input rate, and tool calling only works through the Responses API.

GPT-6.1 Sol is the model most teams will migrate to this week. OpenAI prices it at $2 per million input tokens and $10 per million output tokens, one fifth of GPT-6 Astra's standard rates. Before you swap the model string and ship, read three lines in its documentation, because each one can quietly change your latency, your bill or your integration.

First, reasoning. Sol supports reasoning.effort values of low, medium, high, xhigh and max, and medium is the default. The none and minimal efforts are not supported. If your code sends minimal to keep a classification or extraction call fast and cheap, that setting does not carry over to Sol. Move those calls to low explicitly, and set the effort on every request rather than inheriting the medium default, which spends more reasoning tokens than simple tasks need.

Second, caching. Cached input on Sol costs $0.10 per million tokens, 5% of the normal input rate. But cache writes cost $2.50 per million tokens, 1.25 times the uncached input rate. Caching pays off handsomely for a long system prompt or document that is reused many times, and it costs you extra for a prefix that is written once and never read again. Put stable content such as instructions, schemas and reference documents at the front of the prompt, keep per request content at the end, and only lean on caching for prefixes you expect to reuse.

Third, the endpoint. OpenAI's documentation says Sol supports Chat Completions without tool calling and directs tool calling to the Responses API. Agent and function calling workloads built on Chat Completions need to move to Responses before they can use Sol's tools.

One more line is worth a sticky note. Prompts with more than 272,000 input tokens are priced at 2x the input and cache rates and 1.5x the output rate for the full request, not just the tokens past the threshold. Sol's large context window makes it tempting to send everything. Trimming a request just under that line can cut its cost substantially.

Try it today

1. Replace any reasoning.effort value of none or minimal with low, and set effort explicitly on every Sol request instead of inheriting the medium default.
2. Reorder prompts so stable instructions and reference material come first and per request content comes last, then cache only prefixes you will reuse, since cache writes cost 1.25 times the input rate.
3. Move any Sol workload that uses tools or function calling from Chat Completions to the Responses API.
4. Log input token counts and flag requests above 272,000 tokens, which are billed at 2x input and 1.5x output for the whole request.
5. Run a sample of current Astra traffic through Sol and compare task success before switching production routes.