GPT-6.1 Sol Explained: OpenAI’s Near-Astra AI Model
 |
| Image: OpenAI |
OpenAI has introduced GPT-6.1 Sol, an upgraded version of GPT-6 Sol designed to deliver substantially stronger performance on coding, computer-use tasks and professional workflows while keeping the same standard API input and output prices as GPT-6 Sol.
OpenAI describes GPT-6.1 Sol as offering near-GPT-6 Astra performance on several demanding workloads, but at a much lower cost. The model became available on September 29, 2026, through the OpenAI API, Codex and ChatGPT Work for eligible plans.
What Is GPT-6.1 Sol?
GPT-6.1 Sol is not a completely new generation after GPT-6. Instead, it is an upgrade to GPT-6 Sol focused on improving performance while maintaining Sol's position between the flagship GPT-6 Astra and the more cost-focused GPT-6 Luna.
OpenAI says the model is particularly aimed at:
- Complex software development
- Agentic coding
- Computer-use workflows
- Professional document analysis
- Multi-step business tasks
- Scientific workflows
- Long-running AI agents
The model supports reasoning levels from low through max, allowing developers to trade off response quality, processing effort and cost depending on the task.
GPT-6.1 Sol vs. GPT-6 Sol
The biggest difference is not the model's basic role in the GPT-6 family. It is the amount of capability OpenAI has added to Sol without increasing its standard API prices.
| GPT-6 Sol | GPT-6.1 Sol | |
|---|---|---|
| Position | Balanced reasoning model | Upgraded balanced reasoning model |
| Focus | Complex work, coding and agents | More demanding coding, agents and professional work |
| Standard input | $2 / 1M tokens | $2 / 1M tokens |
| Cached input | $0.20 / 1M tokens | $0.10 / 1M tokens |
| Output | $10 / 1M tokens | $10 / 1M tokens |
| Context window | Up to 1.05M tokens | Up to 1.05M tokens |
| Max output | Up to 128K tokens | Up to 128K tokens |
The standard input and output prices remain the same, while cached input is now half the GPT-6 Sol price.
Coding Gets a Major Upgrade
Coding is one of the areas where OpenAI reports the largest improvement.
On DeepSWE v1.1, a benchmark for complex software-engineering tasks in real codebases, OpenAI says GPT-6.1 Sol matches GPT-6 Astra while costing roughly one-fifth as much in the tested setup.
It also improves substantially over GPT-6 Sol, with OpenAI reporting a 6.4-percentage-point improvement over GPT-6 Sol's best score in that evaluation.
That makes the model particularly relevant to AI coding agents that need to understand an existing codebase, make changes across multiple files, debug problems and complete longer tasks rather than simply generate short pieces of code.
Better at Using Computers
GPT-6.1 Sol is also designed for tasks where an AI needs to interact with software rather than simply respond with text.
On OSWorld 2.0's offline evaluation, OpenAI reports that GPT-6.1 Sol:
- Scores 7 percentage points higher than GPT-6 Sol at maximum reasoning effort.
- Comes within 2.1 percentage points of GPT-6 Astra.
- Does so at roughly one-seventh of Astra's cost per task in the reported comparison.
These capabilities matter for agents that need to navigate applications, perform multi-step operations and complete workflows on a computer.
Professional Work and Documents
GPT-6.1 Sol also targets professional tasks involving complex information.
On OpenAI's GDP.pdf evaluation, which tests questions involving professional PDF documents containing tables, charts, diagrams and fine-print details, the company reports that GPT-6.1 Sol approaches GPT-6 Astra's performance at approximately one-fifth of the cost per task.
On AutomationBench, which evaluates multi-step business workflows using dozens of tools, GPT-6.1 Sol scored 4.8 percentage points higher than GPT-6 Sol at the same reasoning setting.
This is important because many practical AI applications require more than generating an answer. They require an agent to understand information, use tools and complete several steps correctly.
Scientific Workloads
OpenAI also tested GPT-6.1 Sol on scientific workflows involving data analysis, simulations and theorem proving.
On Terminal-Bench Science 0.1, OpenAI reports that GPT-6.1 Sol more than doubled GPT-6 Sol's score at maximum reasoning effort while costing less than half as much per task.
However, GPT-6 Astra still achieved the highest score in this particular evaluation, so GPT-6.1 Sol should not be interpreted as replacing Astra for every demanding scientific task.
Fewer Factual Errors
OpenAI reports an improvement in factuality as well.
On a deliberately difficult evaluation made from de-identified ChatGPT conversations where users had previously flagged factual errors, GPT-6.1 Sol reduced the percentage of responses containing at least one factual error from 11.4% with GPT-6 Sol to 7.7% at low reasoning effort.
That's an approximately 32% reduction in the measured error rate.
OpenAI cautions that these prompts were specifically selected to expose factual mistakes and are not representative of normal usage.
GPT-6.1 Sol Pricing
The standard API pricing is one of the most notable parts of the release.
GPT-6.1 Sol costs:
- $2 per 1 million input tokens
- $0.10 per 1 million cached input tokens
- $10 per 1 million output tokens
The standard input and output prices are unchanged from GPT-6 Sol, but cached input has dropped from $0.20 to $0.10 per million tokens.
OpenAI says the standard input and output prices are one-fifth of GPT-6 Astra's, making Sol a lower-cost option for applications that need strong reasoning but do not require the flagship model for every task.
GPT-6.1 Sol vs. GPT-6 Astra
The two models occupy different positions in OpenAI's lineup.
GPT-6 Astra remains OpenAI's flagship model for the most demanding reasoning, coding, computer-use and professional workloads.
GPT-6.1 Sol is positioned as the more cost-efficient alternative, with OpenAI reporting near-Astra performance on several evaluations.
That does not mean the two models are identical. OpenAI's own testing still shows Astra leading on some workloads, including the cited scientific research evaluation.
The practical difference is therefore less about replacing Astra and more about making high-level AI capability cheaper to deploy at scale.
Availability
GPT-6.1 Sol is available through the OpenAI API under the model name gpt-6.1-sol.
OpenAI also made it available to eligible Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. At launch, OpenAI stated that GPT-6.1 Sol was not yet available in regular ChatGPT.
OpenAI also announced an upcoming Ultrafast option for GPT-6.1 Sol in Codex, designed to provide substantially faster token generation.
The Bigger Picture
GPT-6.1 Sol shows where OpenAI is taking the GPT-6 family: not simply toward larger or more capable models, but toward different levels of capability and cost for different workloads.
GPT-6 Astra sits at the top for the most demanding work, while GPT-6.1 Sol targets users and developers who want much of that capability at substantially lower cost. GPT-6 Luna remains the more cost-focused option for high-volume workloads.
For developers building AI agents, the improvement could be especially significant. Lower-cost reasoning combined with stronger coding, computer use and tool-based workflows means an agent can potentially handle more steps and longer tasks before the cost becomes a major constraint.
GPT-6.1 Sol is therefore less about being a completely new GPT generation and more about making advanced GPT-6-level capabilities considerably more practical to deploy.
