OpenAI has said it could further lower the price of its latest artificial intelligence model.

On July 15, the South China Morning Post in Hong Kong reported that Sam Altman (샘 알트먼) wrote on X, formerly Twitter, that the flagship model GPT-5.6 Sol already costs about half as much as Anthropic’s Claude Fable5 and could be supplied at one-quarter of the price if needed.

Altman claimed GPT-5.6 Sol is often about twice as token-efficient as Fable when performing the same task. He called GPT-5.6 Sol “half price” and wrote he would be “happy to deliver at one-quarter of the price.” OpenAI unveiled GPT-5.6 Sol last week and set pricing at $5 per 1 million input tokens and $30 per 1 million output tokens. Anthropic’s consumer model Claude Fable5 costs $10 per 1 million input tokens and $50 per 1 million output tokens.

The remarks came as price has emerged as a key competitive factor in the enterprise AI market. Anthropic has increased its presence this year among global enterprise customers and software developers. Competition with OpenAI has intensified, especially as Claude Code has established itself as a major AI coding tool.

GPT-5.6 sol is half the price and ~twice as token efficient as fable in many cases for accomplishing the same task. happy to deliver at one-quarter of the price.

Low-price offensives by Chinese companies are also adding pressure on OpenAI. OpenAI lowered prices for the GPT-5.6 series versus earlier products, but prices for its flagship and mid-tier models remain higher than those of major Chinese rivals. GLM-5.2 from Zhipu AI, a leading Chinese large language model company, charges about $1.40 per 1 million input tokens and $4.40 per 1 million output tokens. DeepSeek’s standard rate for V4-Pro is $0.44 per 1 million input tokens and $0.87 per 1 million output tokens.

China’s big tech companies are also expanding discount competition to secure enterprise customers. ByteDance’s Volcano Engine cloud unit on July 14 announced a promotion offering a 50 percent discount on bundled AI model usage fees through the end of 2026 for users of its enterprise coding platform, Trae. Alibaba, Tencent and MiniMax have also rolled out discounts on flagship models or promotional programs in recent months.

Against this backdrop, users are starting to weigh cost structures as well as performance. Closed models allow access only through paid cloud services, while open-weight models let users download code for free and deploy it on their own equipment. UBS said in a report last month that rising AI token costs are “a real concern for most organizations,” and said this is boosting interest in open-weight models among large global companies.

The pace of performance improvement in Chinese open-weight models is also accelerating. Goldman Sachs said in a report last week that Chinese open-weight models are approaching a “major inflection point in intelligence performance” compared with closed models. As price cuts and performance gains proceed at the same time, total usage costs and deployment methods are becoming more important selection criteria for companies than the advantage of any specific model.

Still, there has also been an assessment that GPT-5.6 Sol has an advantage in efficiency. Artificial Analysis, an independent AI benchmarking platform, said the model posted a higher overall score in coding-agent tasks while cutting output token usage to about one-ninth of DeepSeek V4. Even if the price-cutting competition continues, actual cost per task and token efficiency are increasingly likely to remain key variables in winning enterprise customers rather than simple price lists.