OpenAI has rolled out ChatGPT for Teens, a specialized version of its chatbot tailored for users aged 13 to 17. The new experience is designed with learning-focused tools and additional safety guardrails to address concerns regarding minors' interaction with AI technology. Teen accounts are automatically routed into this environment, which features a study-focused mode, homework nudges intended to discourage answer-copying, quizzes, and visual learning aids.

Product Features and Safety

The platform introduces "Study Hours," a scheduling option that keeps the learning mode enabled by default during selected times. OpenAI framed the design as support for work done outside of school, while maintaining separate offerings such as ChatGPT for Teachers to keep educators in charge of classroom use. An educator highlighted the intent behind the "Responsible Homework Reminder" tool, noting it guides students step-by-step to help them understand mistakes and build confidence rather than simply providing answers.

On safety, teen users receive default protections meant to reduce exposure to harmful or age-inappropriate material. Higher-risk areas covered include self-harm, violence, eating disorders, dangerous activities, and explicit sexual or graphic content. Parents with linked accounts can adjust settings, set quiet hours, and receive limited safety notifications. Added alerts are tied specifically to violence, self-harm, eating disorders, dangerous activities, and explicit sexual or graphic content.

Strategic Partnerships and Projects

OpenAI announced a "signature partnership" with CodeAI, aimed at expanding access to AI tools and learning resources for students and educators. Karim Meghji, CEO of CodeAI, stated that every student should know how AI works, be able to question the technology, catch its mistakes, and know when to stop trusting it.

The company pointed to teen projects built with help from ChatGPT as examples of constructive use. These include WiFind, a search-and-rescue concept using existing Wi-Fi signals to locate people trapped after disasters, and Audemy, an audio-based educational game platform for blind students. OpenAI noted that Audemy reached 200,000 users and featured 50 games.

Regulatory and Legal Context

The launch comes amid growing scrutiny over safeguards for children in the AI sector. Multiple lawsuits and investigations have raised concerns about harmful chatbot interactions, mental health risks, and minors encountering inappropriate content.

Recent legal actions include:

• A federal class-action lawsuit filed by three Tennessee teenagers against Elon Musk’s xAI, alleging that its chatbot Grok was used to create child sexual abuse material by manipulating real photos. A fourth plaintiff has since been added to the case.
• A lawsuit filed by Texas Attorney General Ken Paxton against Discord in April, accusing the company of misleading families while enabling adults to target minors. The investigation began in October 2025 following reports linking the service to the person accused of killing Charlie Kirk, alongside allegations of exposure to sexual exploitation and extremist material.
• A lawsuit filed earlier this year by families of victims in a mass shooting in Tumbler Ridge, British Columbia, against Sam Altman and OpenAI. The plaintiffs claim ChatGPT failed to alert authorities to the incident.
• Multiple lawsuits against OpenAI’s ChatGPT alleging the bot acted as an "unlicensed therapist" or "suicide coach," providing harmful advice or neglecting to refer users to crisis counselors.

A new study challenges the assumption that lower token prices equate to lower overall costs for enterprise AI adoption. Research from AlphaSense, an AI-powered market intelligence platform, found that OpenAI’s GPT-5.6 Sol and Anthropic’s Opus 4.8 outperformed Chinese models Kimi K3 and GLM-5.2 in both cost efficiency and output quality for complex financial analysis tasks.

The findings suggest that enterprises focusing solely on sticker price may incur higher costs when using less capable models that require more processing steps.

Token Price Versus Total Cost

Chinese models appear cheaper at first glance based on per-token pricing. Moonshot charges $15 per one million output tokens for Kimi K3. In comparison, Anthropic charges $25 for Opus 4.8, and OpenAI charges $30 for GPT-5.6 Sol.

However, AlphaSense tested 246 financial analysis tasks, including analyzing earnings transcripts, SEC filings, analyst estimates, and acquisition activity. The results showed that higher capability often leads to fewer tokens and fewer processing steps required to complete a task.

OpenAI’s GPT-5.6 Sol delivered answers with roughly 20% higher quality while costing about 13% less than Kimi K3 on a median basis. Anthropic’s Opus 4.8 generated responses scoring around 13% higher in quality at roughly half the overall cost of Kimi K3.

What the Numbers Show

The data highlights a divergence between unit pricing and total cost of ownership. While Kimi K3 has a 40% lower per-token price than Anthropic’s Opus 4.8 ($15 vs $25), it requires significantly more tokens or steps to achieve inferior results. This inefficiency reverses the cost advantage, making the premium model the more economical choice for knowledge-intensive workloads like financial research.

AlphaSense CEO Jack Kokko noted that some models appearing more expensive based on token price ended up being less costly due to higher efficiency in token usage.

Strategic Implications for Buyers

As enterprises weigh adopting frontier AI models from companies like OpenAI and Anthropic against lower-priced open-weight alternatives from Chinese developers, the metric for evaluation is shifting. AlphaSense argues businesses should evaluate the total cost of completing a task alongside output quality rather than focusing solely on token prices.

The report does not dismiss open models entirely. Companies running AI workloads on their own infrastructure can avoid per-token charges, and less demanding tasks such as email summarization may not require frontier models. AlphaSense suggests a hybrid strategy using multiple models together, employing a routing system to assign different parts of a query to different models. For example, a more capable model might plan a response before handing execution to a smaller, cheaper model.

The broader takeaway is that the AI industry’s next pricing battle may be decided by who delivers the lowest cost per completed task rather than who charges the least per token.