AI model prices halved overnight: what Sol, Luna and Opus 5.5 cost in rupees

OpenAI and Anthropic released new models within hours of each other on Tuesday, and both cut prices rather than raised them.

Anthropic put out Claude Opus 5.5 at $4 per million input tokens and $20 per million output, which it says is about 40 per cent less than Opus 5 on a typical workload. Cache reads dropped to $0.20 per million, a 60 per cent cut. OpenAI answered with two GPT-6 models: Sol at $2 in and $10 out, and Luna at $0.10 in and $0.50 out. Both are priced at half their GPT-5.6 predecessors, with a 90 per cent discount on cache reads.

The same numbers in rupees

At Rs 277.13 to the dollar, per million tokens:

A million tokens is roughly 700,000 to 750,000 words of English. So on Luna, processing the equivalent of a decent-sized book costs less than a plate of biryani.

Why this matters here and not just in San Francisco

Pakistan has one of the largest freelance workforces in the world, concentrated in writing, design, development, data work and customer support. For a freelancer in Lahore or Karachi billing in dollars, model cost is a direct line in the cost of delivery, and it has just been cut roughly in half on the mid tier and by an order of magnitude at the bottom.

The practical effect is on the kind of work that was previously not worth automating. Bulk classification, first-pass transcription cleanup, translation drafts and routine code review were all marginal at the old prices for a small operator. At Rs 28 per million input tokens, they are not.

The flip side is the same arithmetic in your client's hands. Work that is priced by the word or by the ticket, and that a cheap model can do adequately, gets repriced downward by the market. The freelancers who gain are the ones using the tools rather than competing with them on volume.

What to check before you switch

Three things. First, whether the cheaper model actually does your task well enough: the benchmark scores quoted at launch are not your workload. Run your own sample before you move production work.

Second, the cache discount, which is where a lot of the real saving sits if you send the same context repeatedly.

Third, payment. Paying an overseas API bill from Pakistan still means a card that works for international subscriptions, and the exchange rate on the day. Our currency converter prices a dollar bill in rupees, and the dollar rate page has the dated run.

If you are freelancing full time, the tax side has not changed: export of IT and IT-enabled services has its own treatment, and our freelancer tax calculator works out what you owe on foreign-source earnings.