# GPT-6.1 Sol: How it differs from Astra, pricing, and Codex access > GPT-6.1 Sol vs. GPT-6 Astra: API and Codex credit rates, estimated messages on Plus and Standard Business, availability by plan, and model selection. Source: https://ailoot.ai/blog/en/gpt-6-1-sol-pricing/ Canonical: https://ailoot.ai/blog/en/gpt-6-1-sol-pricing/ Language: en Publisher: Ailoot (https://ailoot.ai/) Translation: guide-108 Revision: sha256-8aa7c010657911406bcc42e6dec15fc86405cc52c0d5f5b395137467f00e46fe
Key points GPT-6.1 Sol (`gpt-6.1-sol`) is an OpenAI model that the company describes as delivering results close to GPT-6 Astra at a lower cost. Through the API, it costs $2 per million input tokens and $10 per million output tokens; Astra costs $10 and $50: a fivefold difference, while cached input is ten times cheaper. In Codex and ChatGPT Work on plans with credits, the rates are proportional in the same way: 50 and 250 credits per million input and output tokens, compared with 250 and 1,250, but the amount of work included in the subscription is not determined by the rates themselves. On Plus and Standard Business, OpenAI estimates 15–160 local messages over five hours with Sol, compared with 5–45 with Astra. The model is rolling out in Codex and ChatGPT Work on Plus, Pro, Business, Enterprise, and Edu, but not in regular chat; on Enterprise and Edu, an administrator must enable it first. By default, OpenAI recommends GPT-6.1 Sol in Codex if it is available to your account and client, and Astra for the most demanding work.
[Ailoot documentation: API, keys, and connection](https://ailoot.ai/docs/). ## What is GPT-6.1 Sol GPT-6.1 Sol is a new OpenAI model for complex work with code, applications, and documents. Its identifier is `gpt-6.1-sol`. On the model page, OpenAI explains why it is needed:

GPT-6.1 Sol delivers near-Astra performance at a lower cost for complex coding, computer use, and professional work. Compare it with Astra on your tasks to assess the tradeoff between quality and cost.

— OpenAI, https://developers.openai.com/api/docs/models/gpt-6.1-sol
Translation: GPT-6.1 Sol delivers results close to Astra at a lower cost for complex coding, computer use, and professional tasks. Compare it with Astra on your own tasks to assess the tradeoff between quality and cost. “Near-Astra” means “close to Astra,” not “the same”: it is a separate model, and OpenAI itself suggests checking the difference on your own tasks. Sol and Astra have the same context and token limits: a context of 1,050,000 tokens, up to 922,000 input tokens, and up to 128,000 output tokens. Both models have knowledge through April 30, 2026, and accept text and images as input. They also have the same set of reasoning levels in the API: `low`, `medium`, `high`, `xhigh`, and `max`. For Sol, OpenAI separately specifies that the default is `medium`, and that the `none` and `minimal` levels are not supported. ## How does GPT-6.1 Sol differ from Astra? OpenAI calls GPT-6 Astra its most capable model and recommends it for the most demanding work: complex reasoning, coding, computer use, research, and document creation. The Codex model selection recommendations say:

Choose Astra when a task needs the strongest capability across steps and tools. GPT-6.1 Sol offers near-Astra performance at a lower cost than Astra. Use GPT-6.1 Sol when available to your account and client, and Luna for clear, repeatable tasks.

— OpenAI, https://learn.chatgpt.com/docs/models
Translation: choose Astra when you need the strongest model for multistep work with tools. GPT-6.1 Sol delivers results close to Astra at a lower cost. Use GPT-6.1 Sol if it is available to your account and client, and Luna for clear, repeatable tasks. OpenAI recommends different models for different tasks: - **Astra** — for the most complex work from start to finish: long sequences of actions with code, applications, and research that require step-by-step reasoning and careful decisions. According to OpenAI, Astra asks clarifying questions better and takes the original goal and constraints into account. - **GPT-6.1 Sol** — for repetitive, extended work with code, applications, and documents when cost matters. OpenAI recommends reserving the most demanding tasks for Astra. - **Luna** — for clear, repetitive tasks at scale: extracting data, classifying, transforming, and assembling a structured summary. In its API model selection guide, OpenAI gives examples where Sol is worth considering: preparing a board presentation based on financial results or building a website from a product description. To evaluate quality and cost, the company recommends comparing it with Astra on the same task. If you build your own projects with AI agents, check your connection setup against the documentation. ## How much does GPT-6.1 Sol cost The cost depends on how you work: through a ChatGPT subscription or the API. - **ChatGPT subscription** (Plus, Pro, Business, and others). You work in Codex or ChatGPT Work and use your plan's limits. When the included limits run out, available credits let you continue; on Plus and Pro, you can buy more without changing your plan. - **OpenAI API with a key.** Usage is charged at the API rates per million tokens. OpenAI separately warns that API prices cannot be used to estimate how many tasks are included in a subscription:

API token prices are separate from subscription usage; don't use them to estimate included tasks.

— OpenAI, https://learn.chatgpt.com/docs/pricing
Translation: API token prices are separate from subscription usage; do not use them to estimate how many tasks are included in the plan. ### API pricing Prices per million tokens in standard mode, according to the model pages as of October 1, 2026: | | GPT-6.1 Sol | GPT-6 Astra | How many times cheaper Sol is | |-------------------|-------------|-------------|-------------------------------| | Input | $2 | $10 | 5 times | | Cached input | $0.10 | $1 | 10 times | | Cache write | $2.50 | $12.50 | 5 times | | Output | $10 | $50 | 5 times | The pricing rules for both models are largely the same. If the input exceeds 272,000 tokens, the doubled input and cache rates and the 1.5-times output rate apply to the entire request. Fast mode costs twice as much as standard mode; Batch and Flex cost half as much. The Sol page separately states that regional processing adds 10% where available; there is no such line on the Astra page. ### Usage in Codex and ChatGPT Work subscriptions ChatGPT Work and Codex use shared limits under the same rules. For plans with credits, OpenAI publishes rates in credits per million tokens in standard mode: | Credits per 1 million tokens | Input | Cached input | Output | |------------------------------|-------|--------------|--------| | GPT-6 Astra | 250 | 25 | 1,250 | | GPT-6.1 Sol | 50 | 2.5 | 250 | | GPT-6 Sol | 50 | 5 | 250 | | GPT-6 Luna | 2.5 | 0.25 | 12.5 | The ratio between GPT-6.1 Sol and Astra rates is the same here: input and output are five times cheaper, and cached input is ten times cheaper. Codex credits have no separate cache write charge. Compared with GPT-6 Sol, the new version differs in rates only for cached input: 2.5 credits instead of 5. OpenAI warns, however, that credit rates alone do not determine how much work is included in a subscription. Separately from the rates, OpenAI publishes estimates of the number of local messages over five hours within the Plus and Standard Business limits: | Model | Plus | Standard Business | |-------------|--------|-------------------| | GPT-6 Astra | 5–45 | 5–45 | | GPT-6.1 Sol | 15–160 | 15–160 | | GPT-6 Sol | 15–150 | 15–150 | These are OpenAI estimates, not fixed limits: the usage page shows your current limits and when they reset. The range is wide because usage depends on project size, context, reasoning, and tools: a small routine function may use only a small fraction of the limit, while a long session with a large context can use considerably more. Cloud tasks may use more than local messages. As of October 1, 2026, Pro has no five-hour limit. ## Who can use it and how to choose At launch, GPT-6.1 Sol is rolling out on Plus, Pro, Business, Enterprise, and Edu plans: in Codex in the ChatGPT desktop app and CLI, and in ChatGPT Work in a browser and on a phone. The restrictions are: - **Free and Go** are not included at launch. - **Enterprise and Edu**: the model is turned off until a workspace administrator enables it. - **Regular ChatGPT chat**: GPT-6.1 Sol is available in Work and Codex, but not in Chat. - **Speed modes**: Standard and Fast are available at launch; Ultrafast support will come later. In an interactive CLI session, enter `/model` and choose a model from the list. At launch, set the model with the `--model` flag or its short form, `-m`: ```bash codex --model gpt-6.1-sol ``` The same flag works in non-interactive mode: ```bash codex exec -m gpt-6.1-sol "Review the current changes" ``` To make the model the default, set it in `config.toml`, the Codex settings file. It is shared by the ChatGPT desktop app, Codex CLI, and IDE extension: ```toml model = "gpt-6.1-sol" ``` In the ChatGPT desktop app and ChatGPT Work in a browser, start with the Power setting available to your account by default. Choose a specific model, reasoning level, or speed in the **Advanced** section. Choosing a new model does not change workspace permissions or grant access to it. If the model is not yet available on your plan or an administrator has not enabled it, a line in `config.toml` will not help. ## How to compare Sol and Astra on your task To compare the models on your own tasks: 1. Take several tasks you have already done with Astra and can evaluate: a small project improvement, a review of changes, or a document based on a template. 2. Run them on GPT-6.1 Sol with the same descriptions and the same files. Start with the reasoning level set by default in your client. 3. Compare the results: did it do what you asked, how much correction was needed, and were any constraints lost? 4. Check the remaining limits before and after a series of tasks on each model. The usage page in Codex settings on the ChatGPT website shows current limits and reset times; during a CLI session, use the `/status` command. If you are on Pro, there is no five-hour limit, but weekly limits may apply. If so, compare their remaining amounts. Enterprise and Edu with flexible billing have no fixed limits; usage comes from credits. Keep in mind that ChatGPT Work and Codex use a shared limit: if you work in Work during this time, the before-and-after comparison will be inaccurate. If the results on your tasks are satisfactory, you can set Sol as the default for similar work and use Astra for the most demanding tasks. If not, return to Astra: OpenAI recommends it for the most demanding work. ## What you cannot conclude from this “The same Astra, only cheaper” is an inaccurate description. OpenAI says “close to Astra” and reserves Astra for the most demanding work. The price difference is also specific: five times for input and output, and ten times for cached input. Token prices do not translate directly into the number of tasks in a subscription: usage depends on context, reasoning, and tools, and message estimates are given as ranges. The documentation does not say how Sol will perform on your project compared with Astra; that can only be determined by testing it on your own tasks. Prices and terms are as of October 1, 2026; check the official pages before deciding. If you still have GPT-5.5 set in Codex, note that it will be removed from ChatGPT, ChatGPT Work, and Codex on October 14, 2026; this does not affect the OpenAI API. The article [GPT-5.5 will be removed from Codex](https://ailoot.ai/blog/en/codex-gpt-5-5-retirement/) explains what to do in advance. A comparison of Astra with Anthropic's Claude Fable 5.1 is in [GPT-6 Astra or Claude Fable 5.1](https://ailoot.ai/blog/en/gpt-6-astra-vs-claude-fable-5-1/), and the general approach to choosing is in the article [which AI model to choose for a task](https://ailoot.ai/blog/en/ai-model-for-each-task/).
### Sources - [GPT-6.1 Sol: model page (developers.openai.com)](https://developers.openai.com/api/docs/models/gpt-6.1-sol): description, reasoning levels, context, API pricing - [GPT-6 Astra: model page (developers.openai.com)](https://developers.openai.com/api/docs/models/gpt-6-astra): description, context, API pricing - [Codex models (learn.chatgpt.com)](https://learn.chatgpt.com/docs/models): GPT-6.1 Sol availability by plan, model selection, recommendations for Astra, Sol, and Luna - [Codex pricing (learn.chatgpt.com)](https://learn.chatgpt.com/docs/pricing): credit rates, five-hour message estimates, where to check limits - [Model selection (developers.openai.com)](https://developers.openai.com/api/docs/guides/model-selection): when to consider GPT-6.1 Sol
If GPT-6.1 Sol is available to you, run one familiar task on it and compare the result with Astra. If you work in Codex CLI, check your remaining limits with the `/status` command. If you need access to neural networks through Ailoot, see [Ailoot](https://ailoot.ai/). ## Questions about GPT-6.1 Sol: How it differs from Astra, pricing, and Codex access ### Can I calculate my subscription’s task allowance from API token prices? No. API prices and subscription usage are separate. Check your account’s current limits and reset times, and compare models on the same real tasks with the same files and context. ### Will adding gpt-6.1-sol to config.toml unlock it for my account? No. A model setting selects a model only when the account, client, and workspace already have access. Enterprise or Edu availability can also depend on administrator enablement. ## Ailoot · The company and its project portfolio Ailoot combines established operating experience, a large client base, stable service and ongoing KPI control. The company and its project portfolio: millions of customers; monthly sales in the billions. Ailoot brings that operating experience to API access for AI features in customer products, agents, applications and automations.