---
title: OpenAI Launches GPT-6.1 Sol Model
url: https://www.elseif.net/openai-launches-gpt-61-sol-model
published: 2026-10-02T13:04:20+00:00
language: en
section: Models
source: https://www.itmedia.co.jp/aiplus/article/2609/30/2000001867/
organizations: OpenAI, Anthropic
publisher: elseif
---

# OpenAI Launches GPT-6.1 Sol Model

OpenAI announced the release of GPT-6.1 Sol on September 29. This new model arrives only one week after the introduction of its predecessor, GPT-6 Sol.

The company stated that GPT-6.1 Sol offers improved performance across a variety of tasks, including computer operation and coding. While GPT-6 Astra remains the highest performing model from OpenAI, the company noted that GPT-6.1 Sol is better suited for API users building and operating large scale applications, as well as users who wish to perform important tasks more frequently.

In benchmark tests, GPT-6.1 Sol achieved performance equivalent to GPT-6 Astra on the GDP.pdf test for reading PDF documents and the DeepSWE v1.1 test for coding performance, while costing one fifth as much per task. On the OSWorld 2.0 test for computer operation capabilities, the model reduced the score gap with GPT-6 Astra to 2.1 points at approximately one seventh of the cost per task. Additionally, on the AutomationBench test for multi step workflow execution, GPT-6.1 Sol with medium reasoning volume outperformed Claude Opus 5.5 from Anthropic at about one third of the cost.

Regarding safety, OpenAI stated that GPT-6.1 Sol has significantly improved over GPT-6 Sol in alignment evaluations and is approaching the level of GPT-6 Astra. This follows reports from Reuters and The Wall Street Journal that OpenAI postponed the announcement of GPT-6.1 Astra, which was planned for October, due to safety concerns.

GPT-6.1 Sol is available starting the day of the announcement for users of the Business, Enterprise, Edu, Pro, and Plus plans of ChatGPT within Codex and ChatGPT Work. It is not currently available in Chat. The model is also provided via API for developers. The pricing is 2 dollars for input per 1 million tokens, 0.10 dollars for cached input, and 10 dollars for output.