OpenAI has expanded its AI model portfolio with the launch of its GPT-6.1 Sol, a new model for high-end programming, computer and professional applications and at a low price point compared to its more advanced models and their high-end counterparts. Set in between GPT-6 Astra and GPT-6 Luna, the new model can be viewed for users who want to have more sophisticated reasoning and task execution capabilities.
The launch comes at a time when the artificial intelligence industry is already facing questions about the pace and cost of developing more powerful models. OpenAI had unveiled GPT-6.1 Sol in its developer event in its press release at the beginning of this week, highlighting advances in coding, document understanding, factual accuracy, instruction following and safety.
According to OpenAI, GPT-6.1 Sol is able to perform as well as GPT-6 Astra for many demanding professional tasks. It particularly has emphasized coding and multi-step workflows for software development in which users increasingly depend on AI to take the lead in complex work tasks rather than just text.
GPT-6.1 Sol Brings Major Coding Improvements
One of the biggest areas of improvement with GPT-6.1 Sol is software development. OpenAI says that the new model is better at writing and debugging code, understanding complex instructions, and completing tasks that involve several stages.
In the DeepSWE 1.1 coding benchmark, GPT-6.1 Sol performed even better than GPT-6 Astra. However, OpenAI says the model costs about one-fifth as much per task in this particular benchmark. That difference could make the model particularly relevant for developers and organisations that run a lot of AI-powered coding tasks.
The model is also better at working with documents and the workflow of professional work. Rather than simply answering a few questions, GPT-6.1 Sol is intended to handle longer and more complex sequences of work, which could be useful for research, coding, analysis and other knowledge-intensive tasks.
Improved Accuracy And Fewer Factual Errors
OpenAI has also highlighted improvements in factual accuracy with GPT-6.1 Sol. According to the company, the factual error rate at low reasoning effort dropped from 11.4 per cent with the earlier GPT-6 Sol to 7.7 per cent with the new version.
OpenAI also states that in the reasoning settings they tested, GPT-6.1 Sol had an error rate below 1.9 percentage points of GPT-6 Astra. This suggests that the new model is intended to bridge the gap between a more affordable system and OpenAI's higher-end reasoning model.
However, benchmark results do not necessarily represent performance in every real-world situation. AI model performance can vary depending on the type of task, prompting approach, reasoning setting and the information being processed.
OpenAI Focuses On Safety And Instruction Following
Safety and instruction following are another part of the GPT-6.1 Sol upgrade. OpenAI says the model has been improved to follow user instructions more reliably while remaining within its safety restrictions.
The company claims GPT-6.1 Sol is better at detecting problems where tools fail or produce unreliable results. It is also more capable of complying with restrictions and not doing things that have not been authorised.
These capabilities are of particular importance as AI models are increasingly being used as agents able to interact with software and complete multi-step workflows. In these instances, understanding limitations and avoiding unauthorised actions is critical to the performance of the model.
Who Can Access GPT-6.1 Sol?
GPT-6.1 Sol is currently available to Plus, Pro, Business, Enterprise and Edu users via ChatGPT Work and Codex. Developers can also access the model through OpenAI's API by using the model name gpt-6.1-sol.
However, for regular ChatGPT conversations, GPT-6.1 Sol is not available. So access is currently dependent on the product or platform through which the user interacts with OpenAI's models.
OpenAI has also released GPT-6.1 Sol Ultrafast and is releasing another version in the next few days. And the firm said the model could provide up to eight times faster token generation in Codex, which could be useful for developers who value speed in programming.
GPT-6.1 Sol API Price
OpenAI has announced global API pricing for GPT-6.1 Sol at $2 per million input tokens. Cached input tokens are priced at $0.10 per million while output tokens are priced at $10 per million tokens.
The company hasn’t announced a price in India for GPT-6.1 Sol. For developers, the final cost of using the model will depend on the number of input and output tokens consumed, as well as whether cached inputs are used.
With GPT-6.1 Sol, OpenAI is targeting users who want advanced AI capabilities without necessarily paying the higher cost associated with its flagship models. Its coding improvements, lower benchmark cost, stronger factual performance and improved safety would be a great addition to OpenAI's growing model portfolio.
Comments