News Desk, Biggani.org
Just a few years ago, creating a large language model (LLM) with artificial intelligence (AI) was considered a revolutionary technological achievement. Until 2022, building such a model was truly astounding. But fast forward only three years, and by 2025, tech experts are no longer easily impressed by simply building a good model. The real challenge now isn’t just creating a great model, but doing it as cheaply as possible.
In today’s AI world, the main driver of competition has become “how cheaply you can build a language model like GPT or ChatGPT.” Where once building an LLM cost several million dollars, now that cost is being slashed to just hundreds, or in some cases, even a few dollars.
💰 The Six-Million-Dollar Model Is Now Old News
“A six-million-dollar LLM is no longer ‘cool’,” say tech experts. Instead, being able to create a six-dollar model is now a major feat. For example, recent research shows that it’s possible to build a moderately capable language model with limited data and low-power hardware for just a few dollars.
This trend is not driven solely by technological advancement—policy and economic factors are at play too. Whereas AI labs or startups used to require hundreds of millions in funding, now it’s possible to start working on a much smaller scale.
📉 Three Strategies for Cost Reduction
Big organizations like Google DeepMind, OpenAI, and Meta spend heavily to build their language models. But the new generation of innovators are cutting costs using three main strategies—
- Small, targeted datasets: Instead of scraping the entire internet, they’re training on specific topics.
- Budget hardware: Using inexpensive processors instead of costly GPUs.
- Open-source technology: Starting with models from open-source platforms like HuggingFace or Mistral.
This has dramatically reduced costs. For instance, according to one example, at the start of 2025, it became possible to train an LLM for just $500. Imagine that!
🌏 AI for Everyone
The greatest benefit of this cost reduction is that AI is no longer an exclusive tool for the wealthy. Now, even universities, research institutes, or small startups in developing countries like Bangladesh can create their own language models. Building models specifically for the Bangla language is now much easier and more affordable.
We’re seeing small models emerge worldwide for different languages and contexts—specialized language models are being built for African languages, Central Asian regional dialects, and South American indigenous languages.
These models may not be as generalized as GPT-4, but they’re much more effective for their specific contexts.
🧠 Do Smaller Models Mean Less Useful?
Many may wonder—will low-cost LLMs actually work? Researchers respond: “They will, as long as you define your goals clearly.” For example, if a bank develops a small language model just to answer customer inquiries, it doesn’t need the scale of GPT-4. The same applies to village-level agricultural advice, localized medical guidelines, or legal support.
🏗️ A Golden Opportunity for Startups
This reduction in costs has opened doors for small startups to build their own LLMs and use them for business purposes. For instance—
- 🧾 Local language customer service bots
- 👩⚕️ AI health advisory assistants
- 📚 Educational chatbots
- 📈 Analytics tools for local businesses
Where once setting up these services required large amounts of capital, now it’s possible to implement them with investments as low as $500–$1,000.
🇧🇩 What’s the Potential for Bangladesh?
There is some AI work underway in Bangladesh, but the amount of research is still limited. In most cases, foreign models are being used. But if we take advantage of the opportunity to build affordable LLMs, then—
- Custom GPT for the Bangla language
- Specialized models for Chittagong and Sylhet dialects
- Tailored models to address local issues like agriculture, environment, and women’s health
Globally, big AI companies are showing increasing interest in localization. For Bangladesh’s young AI researchers, diving into the world of cheap LLMs is the call of the hour.
🔮 What Does the Future Hold?
If this trend continues, you might soon be able to build your own custom LLM right on your phone—it’ll read your emails, help with daily tasks, and even give personalized career advice.
What was once in the hands of giant Silicon Valley companies, is now becoming ‘technology for everyone’.
✍️ Conclusion
LLMs for a fistful of dollars—while it may sound strange today, this is quickly becoming our future reality. Artificial intelligence is no longer just for big corporations; it’s becoming part of our society, language, business, and daily lives. Technologists, entrepreneurs, and students in Bangladesh should strive to be part of this wave.
The future is waiting—but only for those who are prepared.
📧 Contact: [email protected]
🖊️ Written by: News Desk, Biggani.org
🌐 biggani.org | The Future of Science Stories
Leave a comment