OpenAI Slips In ChatGPT 5.1 As China Steps Up Open-Source Play
OpenAI stresses improvements in coherence and tone control as a lower-cost open-source challenger pushes its case with publicly released benchmarks.
Topics
News
- Parallel Secures $100 Million to Reimagine the Web for AI Agents
- IBM Adds Nighthawk and Loon to Its Quantum Portfolio
- OpenAI Slips In ChatGPT 5.1 As China Steps Up Open-Source Play
- AWS Rolls Out New Tools to Build Job-Ready AI Skills
- Meta’s Chief AI Mind Yann LeCun to Exit, Build Own AI Lab
- Meta Expands AI’s Vocabulary to 1,600 Languages
OpenAI on Thursday released ChatGPT 5.1, three months after its last major update, positioning it as a steadier and more responsive version of its flagship chatbot.
The company said the model handles multi-step instructions more coherently, keeps threads of conversation intact for longer and adapts better when users shift topics midway.
The update splits the model into two main versions. GPT 5.1 Instant is meant for quick, everyday queries, while GPT 5.1 Thinking slows down to handle longer or more complex reasoning.
A new Auto setting decides which version to use without the user having to choose.
Following the launch, OpenAI chief executive Sam Altman posted on X, “I particularly like the improvements in instruction following, and the adaptive thinking. The intelligence and style improvements are good too.”
OpenAI said in a blog that it has made it easier for people to adjust the tone of responses, something users have pushed for because the system often sounded the same across very different conversations.
The company said the aim is to reduce the amount of manual prompting needed to get the desired style.
But the scale of the upgrade has become a point of debate.
Filip Vítek, Executive Vice President for AI and Data at CommentSold, said in a LinkedIn post that many of the changes appear to be tied to how reasoning tokens are allocated.
“It reduces token use for simpler tasks and gives more thinking power to complex ones. The real impact is unclear because there are no official benchmarks from OpenAI,” he said, adding that higher token use on harder queries “will likely push average API usage up.”
“If you want to be cynical, then you also call this token-revenue-maximizing update. As it is clear that higher end of the queries getting 20-70% boost on token burn will take the total weighted average of tokens per query certainly up. So as a result, API usage of GPT 5.1 will be (expectedly) higher,” he added.
The OpenAI update comes as open-source rivals try to close the gap. Moonshot AI in China last week launched Kimi K2 Thinking, which claims stronger performance in agent-focused tasks.
Akshay Pachaar, co-founder of DailyDoseOfDS, also posted on LinkedIn that early tests show Kimi K2 doing well across real-world benchmarks and costing less to run.
Matt Vitale, Executive Director at New Dialogue, said in a post that the model has drawn attention because its weights are public, calling a Chinese open-source model leading reasoning benchmarks “a notable development.”
Tara Tan, founder of Strange Ventures, posted that companies are weighing the economics of running large AI systems and increasingly turning to open-source models for lower costs, faster processing and greater control over their data.
“For operating companies, open models offer lower costs, faster processing, full data control and much more flexibility.This is why businesses like Airbnb are already using models like Alibaba’s Qwen in production,” Tan said.
