appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Minxiao Chang

OpenAI’s latest GPT-5.6 release marks a growing emphasis on cost-efficiency, a move welcomed by some users in China who continue to access the blocked service via virtual private networks (VPNs) and third-party proxies, even as the artificial intelligence model remains more expensive than local rivals.
The launch on Thursday introduced a trio of models tailored to different capabilities, speeds and price points: the flagship Sol, the balanced Terra and the lightweight Luna.
According to OpenAI, Terra delivers performance comparable to its predecessor, GPT-5.5, at roughly half the price, while Luna is engineered to offer robust capabilities at low inference cost.
The release signals a broader industry shift towards cost-efficient AI systems designed for enterprise use, said Li Yitao, a Chinese entrepreneur and co-founder of Canada-based AI start-up Quotaflow.
GPT-5.6 Sol is priced at US$5 per million input tokens and US$30 per million output tokens, while Terra costs US$2.50 and US$15, respectively. Luna, the entry-level one, is priced at US$1 for input tokens and US$6 for output tokens.
Despite the price cuts compared with previous products, OpenAI’s flagship and mid-tier models remain more expensive than major rivals from China. Zhipu AI’s GLM-5.2, for instance, charges about US$1.40 per million input tokens and US$4.40 per million output tokens, while DeepSeek V4 is priced at up to US$0.44 per million input tokens and US$0.87 for output.

But some users have preferred the more costly GPT for its performance.
“ChatGPT has always stood out for its ability to solve problems methodically and balance different considerations, making it particularly suitable for large projects,” Li said.
Li added that enterprise customers were increasingly looking for AI systems capable of making professional judgments, sustaining long, multi-step reasoning and handling complex workflows, rather than simply generating broad, general-purpose responses.
Vincent Liu, a user from Hubei province, said the new model was less likely than GPT-5.5 to drift away from the original objective during multi-turn conversations, allowing users to complete more structured workflows with fewer detailed prompts. “The biggest change I noticed was that the model can now think through a task and expand based on the original framework, rather than being easily guided off course,” Liu said.
GPT-5.6 also performed well in inference efficiency. According to Artificial Analysis, the model completed coding-agent tasks using roughly one ninth of the output tokens consumed by DeepSeek V4 while achieving higher overall scores.
Lower token consumption generally translates into lower inference costs and faster execution, reinforcing OpenAI’s message of delivering more capability with fewer computing resources.
On Chinese social media platform Xiaohongshu, also known as RedNote, a user surnamed Yan praised GPT-5.6’s flexibility on complex tasks, saying the model could proactively use subagents, follow long-term objectives and achieve “extremely high token efficiency”.
On Weibo, user Geek Xiaoyige said GPT-5.6 significantly sped up user interface development tasks, reducing work that previously took hours to around 30 minutes.
OpenAI’s services are not officially available in the Chinese mainland, but local users can access them via VPNs or third-party proxies.
Besides the model, OpenAI also introduced ChatGPT Work, an agent within the chatbot that is able to gather information across apps, create documents and get other work done, expanding beyond just answering questions.


Philippine tribe vows to stand ground against US-led Pax Silica: ‘listen to us’

As the US weighs restrictions on cloud computing, how will China’s AI sector adapt?

Chinese AI chipmaker Hygon plots expansion from data centres to robotics

For China’s hi-tech IPOs, it’s Hong Kong vs the mainland – but how much does it matter?

Artificial intelligence (AI)

Chinese users praise OpenAI’s GPT-5.6  for efficiency, even at higher cost than local rivals

OpenAI on Thursday introduced a trio of models tailored to different capabilities, speeds and price points. Photo: Reuters

Zhipu charges about US$1.40 per million input tokens and US$4.40 per million output tokens for its GLM-5.2 model. Photo: Shutterstock Images

China leads US in everyday AI apps but firms are overvalued, experts say

An ad for Doubao, ByteDance’s AI assistant, is seen at the Shenzhen Bao’an International Airport on May 16, 2026. Photo: Getty Images

The DeepSeek AI assistant app is seen on a smartphone. Photo: Zuma Press Wire/TNS

DeepSeek V4 forces rivals to slash prices, rattling China’s cloud providers

DeepSeek’s ultra-cheap V4 models have triggered a new round of competition, forcing rivals to rethink how they monetise AI. Photo: Reuters

AI unicorn MiniMax on Monday launched its next-generation flagship model, MiniMax M3, pairing token-based billing with subscription plans. Photo: Shutterstock

Tech

News

China Future Tech

The release of the new models signals a broader industry shift towards cost-efficient AI systems designed for enterprise use


The release of the new models signals a broader industry shift towards cost-efficient AI systems designed for enterprise use.


ChatGPT and other generative AIs

OpenAI

China

Weibo

Chinese users praise OpenAI’s GPT-5.6 for efficiency, even at higher cost than local rivals

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}The release of the new models signals a broader industry shift towards cost-efficient AI systems designed for enterprise use

The release of the new models signals a broader industry shift towards cost-efficient AI systems designed for enterprise use