What Is GPT-6.1 Sol? Features, Pricing, and When to Choose It

What is GPT-6.1 Sol, and when should you choose it over Astra? If the growing model catalog makes selection difficult, start with the work you need to finish. GPT-6.1 Sol is a candidate for complex coding and professional tasks where quality and cost both matter. OpenAI describes it as delivering performance close to Astra at a lower cost and recommends comparing the two on your own tasks. That is the provider's positioning, rather than a guarantee of equal results on every workload. Official model page

This guide uses official information checked on October 7, 2026. It explains the specifications, API pricing, differences from the older Sol, and a practical way to evaluate which model fits your work.

Where GPT-6.1 Sol fits

The API model ID is gpt-6.1-sol. OpenAI's changelog records its release on September 29, 2026, for complex coding and professional work at a lower cost than Astra. It also records beta support for Multi-agent. Official changelog

For an evaluation, choose work that involves several constraints: a code fix with test requirements, a synthesis of requirements from documents, or a piece of writing with a specific audience and format. These are suggested evaluation tasks, not measured success stories from this article. Start with an existing task whose output you can inspect and decide whether you would actually use it.

Specifications and settings to check

The model has a 1,050,000-token context window, a maximum output of 128,000 tokens, and an April 30, 2026 knowledge cutoff. It accepts text and image inputs and produces text; audio and video are unsupported. GPT-6.1 Sol specifications

The ability to accept a long document should be evaluated separately from the ability to identify its important details. Begin with relevant passages and ask for a conclusion, the passages supporting it, and unresolved questions. That makes the answer easier to check against the original. For current information, provide or retrieve primary material published after the cutoff and open the cited links yourself.

Supported reasoning.effort values are low, medium, high, xhigh, and max; the default is medium. Neither none nor minimal is supported. Tool calling requires the Responses API, while Chat Completions supports requests without tools. Official GPT-6 guide

API pricing depends on request conditions

For Standard requests with up to 272,000 input tokens, prices per million tokens are USD 2 for input, USD 0.10 for cached input, USD 2.50 for cache writes, and USD 10 for output. Cache reads and cache writes are separate billing categories. Official release and pricing entry

As an arithmetic example, 10,000 uncached input tokens and 2,000 billable output tokens cost USD 0.02 each, or USD 0.04 in total. This is a calculation, not a measured bill. It excludes tool charges, additional billable tokens, and regional conditions. Check actual usage and billing conditions instead of estimating cost from the visible answer's length alone.

Above 272,000 input tokens, input and cache rates double and output rates increase by 50 percent for the entire request. Fast costs twice Standard; Batch and Flex are priced 50 percent below Standard. Regional processing adds a 10 percent premium where available, and tools such as search can incur additional charges. Official pricing conditions

These figures describe metered API use. They do not establish a ChatGPT or Codex subscription price, which models an account can select in an application, or how much usage a subscription includes. Keep API spending and subscription budgeting separate when comparing options.

How Astra and the older Sol differ

OpenAI positions Astra for its most demanding work. Under ordinary Standard conditions, Astra costs USD 10 per million input tokens and USD 50 per million output tokens. GPT-6.1 Sol's corresponding rates are one fifth of those prices. A token-price ratio does not establish the same ratio for task cost or completion time. Astra model page

The older GPT-6 Sol has the same USD 2 input and USD 10 output Standard rates as GPT-6.1 Sol. Its cached input price is USD 0.20, compared with USD 0.10 for the newer model. The older model supports none reasoning and function calling in Chat Completions when reasoning_effort is none. When migrating, check the reasoning setting and endpoint as well as the model ID. Older Sol specifications, GPT-6.1 Sol specifications

Avoid assuming that a newer model is automatically cheaper or that identical ordinary rates imply identical behavior. Existing cache use and tool integrations belong in the migration evaluation.

Compare the models on work you can judge

A useful starting strategy is to try GPT-6.1 Sol on ordinary complex work and compare Astra on tasks where an important requirement is missed. This is a suggested evaluation approach. OpenAI also recommends assessing the quality and cost tradeoff on your own tasks. Official model guidance

Use a repeatable procedure:

1. Select three representative tasks, such as a code repair, a requirements synthesis, and writing with explicit constraints.
2. Define the criteria before running either model: required conditions, traceable evidence, revision count, and which parts you could use.
3. Keep the input, tools, reasoning effort, and processing mode consistent, changing only the model. medium is a convenient initial setting supported by both.
4. Record time to completion and actual usage, including retries and time spent on manual corrections.
5. Repeat the tasks several times and inspect variation and failure types before making a general claim.

For code, check whether the relevant tests pass. For document synthesis, compare the answer with the source's essential requirements. Fluent writing alone can hide additional editing work. Assign tasks that meet your criteria to the less expensive option and separately identify work where accuracy takes priority.

Practical limits before you begin

GPT-6.1 Sol supports US and EU data residency, but Fast is unavailable with EU data residency. A support listing does not establish that your organization meets every eligibility requirement; verify the applicable conditions and configuration. Official model page

Before submitting business documents, follow your organization's information-handling rules and remove unnecessary personal or confidential data. For tool-based tasks, specify the allowed completion boundary, such as proposing a change or saving it. This also makes your evaluation's definition of completion clearer.

This guide synthesizes official documentation and includes no independent performance measurements. It offers no universal speed or success-rate ranking. Use the results of your own representative tasks when deciding what to adopt.

Summary

GPT-6.1 Sol is a candidate when complex work requires a balance of quality and cost. Ordinary API input and output rates match the older Sol, but cached input pricing, reasoning settings, and tool-calling conditions differ. When comparing Astra, inspect revision count and completion time alongside token prices.

To start today, choose one familiar task, write down three essential requirements, and compare Sol and Astra under the same conditions.

As model choices multiply, checking how much editing your own work needs can make the decision clearer than judging a model by its name.