OpenAI Unveils a More Human-Like Chatbot – But Responses May Cost Up to $1,000
OpenAI, the company behind the popular chatbot ChatGPT, has introduced upgraded AI models designed to tackle complex problems with a focus on human-like reasoning and reflection. However, these advancements come with a hefty price tag.
A Leap Forward in AI: The o3 Model
Last Friday, OpenAI launched its latest AI chatbot model, o3, an improvement on its predecessor, o1. The o3 model spends more time reflecting during its response generation, allowing it to solve significantly more complex problems. This reflective process mimics human reasoning and decision-making, marking a major milestone in AI development.
To accommodate diverse needs, OpenAI also introduced a smaller, more cost-effective version called o3-mini. Interestingly, the new model was named o3 instead of o2 out of respect for a British telecommunications brand that uses the latter name.
During OpenAI’s latest livestream event—part of its “12 Days of OpenAI” series—Chief Scientist Mark Chen detailed the o3 model’s enhanced capabilities. The model excels particularly well with intricate questions, making it a powerful tool for users seeking answers to challenging problems.
Performance That Redefines Standards
The o3 model has achieved remarkable results in various fields:
- Mathematics: A near-perfect score of 96.7% on the prestigious American Mathematics Olympiad (AIME).
- Science: Scoring 87.7% in the GPQA Diamond test, which evaluates knowledge in biology, physics, and chemistry at the level of a Ph.D. candidate.
- Coding: Significantly outperforming its predecessor, o1, in programming challenges.
Despite o1 and its smaller counterpart, o1-mini, being released just months ago in September, the o3 models represent a dramatic leap forward.
The Price of Precision
The advanced capabilities of the o3 models come at a cost. Responses generated by the o3 model can exceed $1,000 per question when run on high-end computational infrastructure, according to the AI research platform ArcPrize. The standard version of the model costs between $10 and $100 per task, while o1 answers are priced around $1.
For those curious about AI benchmarks, the o3 model achieves an impressive 87.5% on the ARC-AGI test, which is the gold standard for measuring general intelligence in AI systems. This score surpasses the human average of 85%, although the performance drops to 75.7% with standard computational power.
Aligning AI with Human Values
In addition to technical enhancements, OpenAI aims to align its models more closely with human values, tackling issues like racism and discrimination. “This is no simple task,” explains Jorge De Corte, founder of the Belgian AI company ReBatch. “Humans are not purely rational beings and have subjective characteristics. OpenAI is striving to mimic human thought and behavior as accurately as possible.”
More Innovations on the Horizon
The “12 Days of OpenAI” livestream series unveiled other exciting developments, including:
- ChatGPT Search: A new feature available to all logged-in users, allowing for better search capabilities within the chatbot.
- ChatGPT Pro: Enhanced subscription tiers offering additional features for advanced users.
- Sora: A demonstration of OpenAI’s AI-powered video generation tool.
The o3 models are currently available for testing and research purposes. OpenAI plans to roll out the models to the general public early next year.
For more details on OpenAI’s developments, check out their official blog or visit the 12 Days of OpenAI event page.
The cost of producing a ChatGPT message depends on multiple factors, including the model being used, the computational infrastructure supporting the query, and the complexity of the request. Here’s a breakdown of what goes into the cost:
Key Cost Drivers for a ChatGPT Message
- Model Size and Complexity:
- Larger, more sophisticated models like OpenAI’s o3 model require significantly more computational resources to generate a response. These models involve millions, if not billions, of parameters, making each calculation resource-intensive.
- Infrastructure Costs:
- Hardware: High-performance GPUs or TPUs process the model’s computations.
- Energy Usage: Running large models consumes substantial amounts of electricity.
- Cloud Services: Hosting these services on platforms like Azure or AWS comes with added expenses.
- Prompt Complexity:
- Simple queries, such as “What’s the weather?”, require less computation.
- Lengthy or nuanced tasks, like translating and rephrasing a detailed article, demand more processing power as the model evaluates context, maintains coherence, and generates a structured response.
- Model Variants and Configurations:
- Smaller models (e.g., o3-mini) are more cost-effective than full-scale ones but might not deliver the same level of depth or precision.
Estimated Costs
- High-End Models (e.g., o3):
- Processing one advanced, multi-step query might cost $10–$100.
- Ultra-complex requests or resource-intensive answers can escalate to $1,000+, especially for enterprise-grade queries.
- Standard Models (e.g., ChatGPT 4.0):
- Each message typically costs OpenAI $0.01–$0.05 for standard queries using moderate computational resources.
- Lightweight Models (e.g., older versions or “mini” models):
- Costs drop further, to approximately $0.005–$0.01 per message.
