OpenAI and Anthropic, the companies behind ChatGPT and Claude, have cut the prices of their top AI models by up to 80% since June 2026. The AI price cuts were tracked in a timeline published on October 25, 2026, by Max Woolf, a former senior data scientist at BuzzFeed who writes about AI costs.
The cuts are to the prices developers and businesses pay to plug these models into their own apps and tools. They matter less to people who simply chat with the free apps.
From Rs. 277 to Rs. 55
AI companies charge by the “token,” a small piece of text roughly three-quarters of a word long. Prices are set per million tokens, which is roughly 750,000 words.
The biggest cut came on July 30, 2026. OpenAI dropped the price of its small GPT-5.6 Luna model by 80%, from $1 to $0.20 per million tokens of input. The model had gone on sale only three weeks earlier, on July 9. At the State Bank’s exchange rate of about Rs. 277 per dollar, that is a fall from roughly Rs. 277 to Rs. 55.
Anthropic and OpenAI then cut prices back and forth for the next two months:
| Model | Old price (input / output) | New price (input / output) |
|---|---|---|
| OpenAI GPT-5.6 Luna | $1 / $6 | $0.20 / $1.20 |
| Anthropic Claude Sonnet 5 | was due to rise to $3 / $15 | kept at $2 / $10 |
| OpenAI Sol (GPT-5.6 to GPT-6.1) | $5 / $30 | $2 / $10 |
| Anthropic Claude Opus (5 to 5.5) | $5 / $25 | $4 / $20 |
Prices in US dollars per million tokens.
Anthropic said its new Opus 5.5 needs less computing power to run than Opus 5, and that its price reflects this. Max Woolf estimates that Opus 5.5 costs 40% to 60% less to use than the older model.
Why coding tools gain most
The cuts that matter most are not always the headline prices. AI coding assistants such as OpenAI’s Codex and Anthropic’s Claude Code re-read the whole conversation at every step. Companies charge a much lower “cache” rate for this re-reading.
Max Woolf says these cache reads make up 96% to 99% of what coding assistants spend on tokens. So when Anthropic cut the cache rate on its Fable 5.1 model by 75% in September, the real cost of using it fell by about 70% to 75%. In late September, OpenAI also halved the cache rate on its new GPT-6.1 Sol model.
For Pakistani developers, freelancers and startups who pay for AI in dollars, this means the same work now costs far fewer rupees.
Data from Vercel, a company that routes AI requests for developers, shows the average cost per token has fallen to 43.3% of its April 2026 level. Total spending is still rising, though, because better models are encouraging people to use AI more.
Cheaper rivals force the cuts
Max Woolf argues the cuts are driven by competition from “open-weight” models, whose files are released so anyone can run them. Switching between models is now easy, so customers can move quickly to a cheaper option.
Chinese company DeepSeek’s latest small model, DeepSeek V4.1 Flash 1017, costs just $0.15 per million input tokens. On OpenRouter, a website that lets developers pick between AI providers, 29 companies now offer it. In May, only 13 offered the earlier version.
Moonshot AI’s new Kimi K3.1 Pro scores close to Claude Opus 5.5 on the Artificial Analysis ranking, 55 against 58, at half the input and output price. Moonshot AI plans to release the model’s files on November 4, 2026. Max Woolf expects many providers to start offering it at similar prices.
Even so, these cheaper open models still handle only a minority of all AI usage.
Questions ahead of share sales
The timing is unusual. OpenAI is rumoured to be planning to sell shares to the public in 2027, and Anthropic within the next few months. Leaked financial reports show both companies losing billions of dollars on research and building data centres.
Max Woolf points out that cutting your main source of income by up to 80% is the opposite of what companies usually do before listing. However, some reports put their profit on each token at around 80% before the cuts, which may leave room to lower prices.
Not every change favours users. OpenAI cut the usage allowance on its $200-a-month Codex plan by half, saying its models had become more efficient. Max Woolf himself cut his own Codex plan to one-fifth of its maximum quota, because the cheaper model let him do the same work for less.
He believes prices could keep falling as long as rival providers match the big companies on quality and customers can switch easily.
