OpenAI pitched GPT-6.1 Sol as “near-Astra intelligence for a fifth of the price”: $2 per million input tokens and $10 per million output tokens, against $10 and $50 for GPT-6 Astra. This guide gets you from zero to a first request in curl and Python, then covers the settings and gotchas that decide what your bill looks like.
Key takeaways
- GPT-6.1 Sol costs $2 per million input tokens and $10 per million output, a fifth of GPT-6 Astra.
- Export OPENAI_API_KEY, then call the Responses API with model gpt-6.1-sol from curl or the Python SDK.
- Reasoning effort runs from low to max; hidden reasoning tokens are billed as output, so higher effort costs more.
- Prompts over 272K input tokens cost more, and Sol does not support the none effort or Realtime and embeddings.
Everything here comes from OpenAI’s own docs and launch post, read on 15 November 2026. We checked every command and parameter against those docs and syntax-checked the curl and Python samples (the JSON body parses and the Python compiles). Live calls need your own API key and are billed, so the sample responses shown are OpenAI’s.
What you need
- An OpenAI API key. Create one in the dashboard at platform.openai.com/api-keys and keep it out of your code.
- Billing set up. The quickstart points you to the billing page to add credits.
- Python 3.10 or newer for the Python route, according to the SDK’s README, plus
pip. - curl for the command-line route.
- The model ID:
gpt-6.1-sol, per OpenAI’s model page.
Step 1: Export your key
On macOS or Linux:
export OPENAI_API_KEY="your_api_key_here"
On Windows, the quickstart uses setx OPENAI_API_KEY "your_api_key_here". The official SDKs read the key from this environment variable automatically.
Step 2: Make your first request with curl
OpenAI’s quickstart makes its first request through the Responses API. The docs’ examples use gpt-6-astra, so we have swapped in Sol’s model ID.
curl https://api.openai.com/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"model": "gpt-6.1-sol",
"reasoning": {"effort": "low"},
"input": "Explain what an API key is in two sentences."
}'
Expected result: a JSON response. The generated text sits inside the output array, in an item whose type is message.
Step 3: Make the same request in Python
Install the official SDK:
pip install openai
Save this as example.py:
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-6.1-sol",
reasoning={"effort": "low"},
input="Explain what an API key is in two sentences.",
)
print(response.output_text)
Run it with python example.py. Expected result: a short answer printed in your terminal. The SDK’s output_text property saves you from digging through the JSON.

Step 4: Pick a reasoning effort
The reasoning.effort setting controls how long the model thinks. For GPT-6.1 Sol, the model page lists low, medium (the default), high, xhigh and max. Lower effort favors speed and fewer tokens, according to the reasoning guide.
Reasoning tokens are hidden, but they use context space and are billed as output tokens. Higher effort can therefore cost more even when the visible answer is short.
What Sol can and cannot do
| Detail | Per OpenAI’s model page |
|---|---|
| Context window | 1,050,000 tokens |
| Max output | 128,000 tokens |
| Knowledge cutoff | 30 April 2026 |
| Input | Text and images |
| Output | Text only |
| Endpoints | Responses, Chat Completions, Batch |
| Endpoints not supported | Realtime, Assistants, fine-tuning, embeddings, image generation, audio |
Tool calling requires the Responses API. If you also build against Google’s models, our guide to the Gemini 4 Hydrogen API covers the same ground in Python, JavaScript and curl.
By the numbers
1,050,000 tokenscontext window
128,000 tokensmaximum output
30 April 2026knowledge cutoff
What it costs
Standard prices per million tokens, from OpenAI’s launch post of 29 September 2026. They matched OpenAI’s pricing page when we checked on 15 November:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
The model page adds the fine print:
- Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request.
- Fast mode costs 2x Standard, and Batch and Flex are 50% lower.
- Ultrafast mode costs 6x Standard. It needs
service_tierset toultrafast, and OpenAI strongly recommends WebSockets for it.
A quick sum using the standard rates: 10,000 input tokens and 1,000 output tokens come to about $0.03, before any reasoning tokens. That is our arithmetic, not an OpenAI example.

Common errors and fixes
OpenAI’s docs do not list exact error messages for these, so treat them as the causes to check first.
- Authentication fails. Check that
OPENAI_API_KEYis set in the same terminal session you are running the command from. - Sampling settings cause trouble. When reasoning effort is not
none, the migration notes say to removetemperature,top_pandtop_logprobs. - You asked for
none. GPT-6.1 Sol does not support thenoneeffort. Uselowinstead. - Tools misbehave in Chat Completions. The docs say tool calling requires the Responses API.
- The wrong endpoint. The Embeddings, image generation, audio and Realtime endpoints are not supported for this model.
- The bill looks too high. Check for prompts above 272K tokens and for high reasoning effort.
- You cannot find Sol in ChatGPT chat. At launch on 29 September 2026, OpenAI said it was available in ChatGPT Work and Codex but not yet in Chat. Check the release notes for current status.
Did you know?
Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output, and that applies to the full request.
When Sol is the right pick
OpenAI’s guide says to use Astra for the most demanding work, Sol for complex work at lower cost, Terra for visual reasoning and writing, and Luna for focused, high-volume tasks. For one example, the launch post says Astra still posts the highest score among the models OpenAI tested on Terminal-Bench Science 0.1, at 68.1%, and OpenAI says it should be used for the hardest scientific research tasks.
If you only want to experiment without paying per token, you can run a model on your own computer with Ollama. For the production route, start with low effort and raise it only when quality demands it.
Frequently asked questions
How much does the GPT-6.1 Sol API cost?
Standard prices are $2.00 per million input tokens, $0.10 per million cached input tokens and $10.00 per million output tokens. Batch and Flex are 50% lower, Fast mode costs 2x and Ultrafast 6x Standard.
What is the model ID for GPT-6.1 Sol?
The model ID is gpt-6.1-sol, per OpenAI’s model page. You pass it as the model value in a Responses API request, from curl or the official Python SDK.
Does GPT-6.1 Sol support the none reasoning effort?
No. The model page lists low, medium (the default), high, xhigh and max. If you ask for none, use low instead, which favors speed and fewer tokens.
Can GPT-6.1 Sol generate images or audio?
No. Sol takes text and images as input but outputs text only, and the image generation, audio, Realtime, embeddings, Assistants and fine-tuning endpoints are not supported for this model.
Quick quiz
Which reasoning effort is the default for GPT-6.1 Sol?
- Low
- Medium
- Max
The model page lists medium as the default effort.
How are hidden reasoning tokens billed?
- Not billed
- As input tokens
- As output tokens
Reasoning tokens are hidden but use context space and are billed as output tokens.
Which API is required for tool calling with Sol?
- The Responses API
- The Assistants API
- The Realtime API
The docs say tool calling requires the Responses API.
Sources
Links last checked on 15 November 2026.





Write a response