Two things happened in the AI market this week that matter more to a small business in India than any model launch this year. OpenAI released GPT-6.1 Terra at half the API price of its predecessor. And IBM's CEO told investors that open-weight models now deliver 90% of frontier-model effectiveness at 10-25% of the token cost.
Put together: the price of useful AI is collapsing. That changes the maths for every business that looked at AI automation in the last year and decided it didn't pay back.
What was actually announced
On October 26, OpenAI introduced GPT-6.1 Terra, a model built for practical work: reading documents, filling forms, operating software, drafting proposals. The numbers from OpenAI's own announcement:
- Standard API pricing is 50% lower than GPT-5.6 Terra - $1 per million input tokens, $5 per million output tokens.
- It matches the flagship GPT-6 Astra on professional task benchmarks at one-tenth the price per token, and works out to roughly 9x lower cost per task than Anthropic's Claude Opus 5.5 at maximum effort.
- Factual error rates fell from 53.6% to 16.7% at low effort (and to 4.3% at maximum effort).
Days earlier, IBM CEO Arvind Krishna made the case from the other side on his company's earnings call. His framing of open-weight models: "Would you like something that's 10 percent of the cost if it's 90 percent as good as the best, or do you want to pay that 10 times higher cost?" His list of tasks that don't need the expensive model is worth quoting because it describes most business work: "summarizing an email, examining a contract, answering a customer call, composing an email."
Why the timing matters: the cost side of the ROI question
Just days ago, Apollo's chief economist Torsten Slok made headlines arguing there is still no measurable AI payoff outside the tech sector - flat revenue per worker and margins across most of the market. We covered what that means for Indian businesses. The fair reading of the two stories together: the payoff question has two sides, what AI earns you and what AI costs you. The earnings side is still unproven for most companies. But the cost side just moved dramatically in your favour.
Automations that were marginal at last year's prices - reading incoming invoices into accounting software, drafting first-pass replies to customer enquiries, summarising contracts before a human reviews them - clear the bar much more easily at one-tenth the cost.
What Indian small businesses should actually do
1. Revisit the ideas you shelved as "too expensive"
Any workflow you evaluated six months ago and rejected on cost deserves a second look. The price of the AI component may have fallen by half or more since then.
2. Match the model to the task, not the brand
Krishna's point applies doubly in India, where budgets are tighter: the daily work of a business - summaries, drafts, data entry, first-pass customer replies - does not need the most expensive model. Reserve premium models for the tasks that genuinely require them.
3. Keep a human on the output
The honesty in OpenAI's own numbers deserves attention: even after a dramatic improvement, the new model still makes factual errors on 16.7% of error-inducing prompts at low effort. Cheap AI that nobody checks is not cheap. Build the review step into the workflow from day one.
4. Don't self-host unless you have the people
Krishna was candid about the catch with cheap open-weight models: running them yourself takes engineers who understand GPUs, networking and caching. For most Indian SMBs, the practical route is still the API - just at the new, lower prices. On-prem AI is a serious option for larger organisations with data-residency needs, not a weekend project.
The honest catch
Falling prices do not create a payoff by themselves. They remove one excuse. The businesses that will actually benefit are the ones that pick a specific, measurable workflow - invoice processing, enquiry response time, contract turnaround - and wire cheap AI into it with a before-and-after number they track. "We use AI" is not a result. "Invoices that took 4 hours now take 40 minutes" is.
The bottom line
AI's cost per task is falling faster than almost any business input in memory. The winners won't be the companies that bought the most expensive model - they'll be the ones that matched cheap, capable models to the right everyday tasks, with human review where it counts.
If you want an honest assessment of which of your workflows would actually pay back at today's AI prices, talk to us for a free consultation.
