The AI Pricing War Just Got Confusing. Here's What Actually Changed
If you've been using ChatGPT, Claude, DeepSeek, or Grok this month, your bill, or your free usage, probably shifted without you noticing. In under three weeks, four of the biggest names in AI moved on pricing. And they didn't move in the same direction.
Two cut prices sharply. One raised prices by over 1,000% on its budget tier. One kept the sticker price the same but added a rule that can quietly double your bill.
There's a reason two of the biggest US labs cut prices at the same time: Chinese AI labs have gotten good enough, fast enough, that companies are actually running the numbers and switching. One widely cited benchmark put the cost of a task at roughly $544 on a Chinese model versus around $4,811 for the same job on a leading US model, a gap large enough that finance teams can't ignore it anymore. That pressure is a big part of why OpenAI and Anthropic both moved this month.
Here's what happened, and what it actually means for you.
OpenAI made its cheapest model free, and unlimited
Then OpenAI went further. Starting the week of August 10, Luna became the default model for anyone on ChatGPT's free and Go tiers, with unlimited text chats. That's a real shift: free users now get an AI model with no message cap, at least for regular text conversations. Image generation and voice mode aren't covered by the unlimited claim, and OpenAI hasn't spelled out limits there yet.
If you already pay for ChatGPT Plus, Pro, Business, or Enterprise, your subscription price hasn't changed. You'll just notice your usage allowance stretches further, since Luna and Terra now cost less per task behind the scenes.
What this means for you: if you're a casual or free ChatGPT user, you just got a meaningfully better deal without doing anything. If you build on OpenAI's API, Luna is now one of the cheapest capable models on the market.
Anthropic cut its flagship price in half, too
Anthropic didn't sit still either. The company introduced Claude Opus 5, priced at roughly half of what its previous flagship, Fable 5, cost, while also scrapping a planned price increase on its mid-tier Sonnet 5 model. Anthropic is pitching Opus 5 as frontier-level performance at that lower price, which only makes sense if the company expects a lot more people to use it now that it costs less.
This wasn't a coincidence. It landed within days of OpenAI's cut, and both moves trace back to the same pressure: Chinese labs have closed the capability gap enough that switching no longer means giving up much quality, just paying a lot less. OpenAI's Sam Altman said as much publicly, noting the company would go even lower on price if competition demanded it.
What this means for you: if cost has kept you from using Anthropic's top-tier model for everyday work, that math just changed. Worth revisiting if you wrote it off as too expensive even a few months ago.
DeepSeek just made itself a lot more expensive
The company also introduced peak and off-peak pricing for the first time. Peak hours are 1–4 AM and 6–10 AM UTC. Outside those windows, you pay half the peak rate.
The numbers: V4-Flash output tokens jump from a flat $0.28 per million to $1.32 per million at peak, or $0.66 off-peak. V4-Pro output goes from $0.87 per million to $3.96 at peak, more than four times its old price.
DeepSeek says the new structure is meant to spread demand more evenly across the day, pushing developers toward quieter hours. It's a fair point, the company's models became so popular that server capacity is clearly a real constraint now.
Even with the increase, DeepSeek is still cheaper than most Western rivals. Anthropic's top model charges $50 per million output tokens, for comparison. But the gap that made DeepSeek the obvious budget pick has narrowed. At peak hours, DeepSeek's V4-Flash output now costs more than OpenAI's newly discounted Luna. Off-peak, DeepSeek still wins, but by a smaller margin than before.
What this means for you: if you built a workflow around DeepSeek's old flat rate, check your numbers again, especially if your usage tends to land during those peak windows. Off-peak scheduling just became a real cost-saving strategy, not just a nice-to-have.
Grok 4.6 didn't raise prices, but it added a trap
Here's the part that's easy to miss. Grok 4.6 has a 500,000-token context window, which sounds generous. But once a single prompt crosses 200,000 tokens, the entire request, not just the extra tokens, gets billed at double the rate: $4 per million input, $12 per million output.
That's a meaningful difference from how most pricing tiers work. Usually, when you cross a threshold, only the tokens above that line cost more. With Grok 4.6, going one token over 200,000 reprices the whole conversation.
What this means for you: if you're using Grok for quick tasks, nothing changes. But if you're feeding it long documents, full codebases, or long meeting transcripts, keep an eye on where you land relative to that 200,000-token line. It's the kind of detail that can double a bill without any warning on the screen.
Quick Comparison Table
| Model | What changed | Effective date | Watch out for |
|---|---|---|---|
| GPT-5.6 Luna (OpenAI) | Price cut 80%, now free default for unlimited text chat | July 30 / Aug 10, 2026 | Free tier covers text chat only, not images or voice |
| Claude Opus 5 (Anthropic) | Launched at roughly half the price of prior flagship Fable 5 | Aug 2026 | Still a premium-tier model despite the cut, check it against your budget model of choice |
| DeepSeek V4-Flash / V4-Pro | Prices up 50–1,100%, new peak/off-peak billing | Aug 16, 2026 | Peak-hour output can cost more than OpenAI's Luna |
| Grok 4.6 (xAI) | Sticker price unchanged | Aug 12, 2026 | Entire request re-priced once you cross 200K tokens |
The Takeaway
Cheap AI used to mean picking the model with the lowest number on the pricing page and sticking with it. That's not really true anymore. The cheapest option now depends on what time of day you're working, how long your prompts are, and which tier you're actually on. A model that was the obvious budget pick in July might not be the cheapest one in your specific workflow by September.
If you're choosing between tools right now, it's worth running your own usage pattern against these numbers rather than going by reputation. What was true about "the cheap one" a month ago may already be out of date, and given how often these companies have moved this year, it probably will be again soon.
Tracking AI Pricing Changes
With pricing updates happening weekly across major AI platforms, it's critical to stay informed before committing to a tool. Alternates.ai compares real-time pricing across all major AI models and platforms, so you can make cost-aware decisions based on your actual usage patterns. Browse Alternates.ai's AI assistant category to benchmark current pricing and find the best fit for your workflow.