OpenAI Shakes Up the AI Market with Dramatic Price Cuts and Performance Fixes
OpenAI has delivered a one-two punch to the AI industry, slashing prices across its model lineup while addressing the notorious "laziness" issues plaguing its flagship GPT-4 Turbo. The moves signal a major shift in how AI companies price their services and respond to user feedback.
The price reductions are staggering, with GPT-4o costs dropping by 83% from earlier GPT-4 pricing levels. Meanwhile, the company has rolled out fixes for GPT-4 Turbo's tendency to incompletely execute tasks, a problem that had frustrated developers across Asia and beyond.
Massive Price Drops Reshape AI Economics
The new pricing structure puts advanced AI capabilities within reach of smaller developers and startups across Asia. GPT-4o now costs $2.50 to $5.00 per million input tokens✦, down from the original GPT-4's $30.00 per million tokens.
This pricing revolution extends beyond OpenAI, with industry-wide API✦ costs averaging $2.50 per million tokens as of early 2025. The trend mirrors broader developments in how AI is reshaping consumer behaviour across Asia, making sophisticated tools accessible to a wider range of users.
GPT-4o mini offers even more aggressive savings at $0.15 to $0.30 per million input tokens. The 60% reduction from GPT-3.5 Turbo pricing makes it particularly attractive for high-volume applications like customer service chatbots.
By The Numbers
- GPT-4o pricing dropped 83% from earlier GPT-4 levels to $2.50-$5.00 per million input tokens
- GPT-4o mini costs 60% less than GPT-3.5 Turbo at $0.15-$0.30 per million input tokens
- Industry-wide average API costs fell 75% to $2.50 per million tokens in early 2025
- Original GPT-4 launched at $30.00 per million input tokens with 8K context window✦
- New GPT-5.1 model priced at $1.25 per million input tokens for latest capabilities
Tackling the "Lazy" GPT-4 Problem
OpenAI's latest update addresses a peculiar issue where GPT-4 Turbo would start tasks but fail to complete them fully. Developers reported instances where the model would begin writing code or analysing data, then abruptly stop mid-task.
The new GPT-4 Turbo version 0125 includes specific fixes for this "laziness" behaviour. Early testing suggests the model now maintains consistency throughout longer tasks, crucial for applications in AI-powered healthcare solutions where completeness matters.
"Our goal is to make AI more accessible and ensure its responsible development. These updates reflect our commitment to that mission," said Mira Murati, OpenAI's Chief Technology Officer.
The timing coincides with OpenAI's broader expansion plans, including new operations in Singapore that will bring these improved models closer to Asian developers.
Vision Capabilities and Content Moderation Upgrades
GPT-4 Turbo with vision represents OpenAI's push into multimodal✦ AI applications. The model can process images alongside text, opening possibilities for visual analysis in fields ranging from medical diagnosis to quality control manufacturing.
The company has also released an updated moderation API (version 007) designed to better identify harmful content. This improvement addresses growing concerns about AI safety✦, particularly relevant given recent developments in AI reasoning models.
| Model | Input Cost (per million tokens) | Output Cost (per million tokens) | Key Features |
|---|---|---|---|
| GPT-4o | $2.50-$5.00 | $10.00-$15.00 | Flagship model, 83% price cut |
| GPT-4o mini | $0.15-$0.30 | $0.60-$1.20 | Lightweight, 60% cheaper than GPT-3.5 |
| GPT-5.1 | $1.25 | $10.00 | Latest generation model |
| GPT-5-pro | $15.00 | $120.00 | Premium tier for advanced tasks |
Subscription Models Under Review
OpenAI faces a strategic dilemma with its unlimited ChatGPT subscription plans as AI costs continue rising. The company is reconsidering whether flat-rate pricing remains viable for resource-intensive models.
"There's no world in which pricing doesn't significantly evolve. It's possible that in the current era, having an unlimited plan is like having an unlimited electricity plan. It just doesn't make sense," explained Nick Turley, an OpenAI executive.
This uncertainty reflects broader industry challenges as AI capabilities advance. The shift could particularly impact enterprise AI adoption patterns across Asia, where many organisations rely on predictable pricing models.
Several factors drive the pricing evolution:
- Computational costs remain high for advanced models despite efficiency improvements
- Growing user demand strains infrastructure capacity across regions
- Competition from other AI providers creates pressure for sustainable pricing
- Enterprise customers require predictable costs for budget planning
- Resource allocation becomes critical as model capabilities expand
Impact on Asia's AI Development
The price reductions could accelerate AI adoption across Southeast Asia, where cost sensitivity remains a significant barrier. Startups in markets like Indonesia, Thailand, and Vietnam now have access to capabilities previously reserved for well-funded companies.
Regional developers can leverage✦ these tools for localised applications, from language tutoring platforms to healthcare chatbots designed for specific cultural contexts. The democratisation of AI access aligns with growing interest in AI companion technologies across the region.
How significant are these price reductions for developers?
The 83% price drop for GPT-4o represents a fundamental shift in AI economics, making advanced capabilities accessible to individual developers and small teams for the first time. This could trigger a wave of innovation across Asia's startup ecosystem✦.
What caused the "laziness" issue in GPT-4 Turbo?
The exact technical cause hasn't been disclosed, but the issue appeared related to context management in longer tasks. Version 0125 includes architectural improvements that maintain model attention throughout extended operations, ensuring task completion.
Will unlimited ChatGPT subscriptions disappear?
OpenAI is actively reviewing its subscription model sustainability. While no timeline exists for changes, the company suggests that usage-based pricing may become necessary as computational costs for advanced AI remain high.
How does GPT-4 Turbo with vision work?
The multimodal model processes both text and images simultaneously, enabling applications like visual document analysis, medical image interpretation, and quality control automation. It maintains the same pricing structure as standard GPT-4 Turbo.
What does this mean for AI competition in Asia?
OpenAI's aggressive pricing could pressure regional competitors to reduce costs or differentiate through specialised capabilities. This competitive dynamic benefits Asian businesses seeking AI solutions but may challenge local AI startups.
The confluence of lower prices, improved performance, and enhanced capabilities positions OpenAI's models as increasingly attractive options for Asian developers and enterprises. As the company continues expanding its regional presence, these updates could reshape how AI gets integrated into daily life across the continent.
What applications will you build with OpenAI's more affordable models, and how might the improved GPT-4 Turbo change your development workflow? Drop your take in the comments below.
Latest Comments (4)
i'm still trying to understand the 'laziness' issue with GPT-4 Turbo. was it really lazy or just hitting some kind of token limit? can someone explain how that actually works under the hood?
Ah, the "laziness" in GPT-4 Turbo. En effet, we've seen similar patterns in some of our RL agents during complex task sequencing. Important to understand the root cause there, not just patch it.
The GPT-3.5 Turbo price drop is good but for us in Malaysia, connectivity and data costs are often bigger hurdles for widespread AI adoption than just the model API fees. Making the models cheaper doesn't automatically mean everyone can use them easily if internet access is still a bottleneck in some areas.
the price drop for GPT-3.5 Turbo is good for the small dev shops but for us, building compliance automation, the cost was never the biggest hurdle for GPT-4. it's always been about predictability and explainability for regulators. the "laziness" fix for GPT-4 turbo is interesting, maybe finally we can get consistent output?
Leave a Comment