The AI Price War Just Got Serious. Here's What It Means For You.

Last week, Anthropic launched Claude Sonnet 5 and turned it into the default free model. That was the story. This week, three more labs shipped competitive frontier models in the space of about 90 hours. If you have not been paying attention to the AI news cycle, this is the week to start. Here is what actually happened, and the one honest question you should be asking yourself as a small business owner.

1. Meta officially entered the AI API market

Meta launched Muse Spark 1.1 on July 9 through its brand-new Meta Model API. It is Meta's first-ever paid model, and its first public API of any kind. Multimodal, agentic, 1-million-token context window. Priced at $1.25 per million input tokens. And every new account gets $20 of free credits to test with.

Translation: the market just gained a new billion-dollar competitor with a real free trial. If you have never touched Meta AI because you assumed it was locked inside Facebook and Instagram, that assumption is out of date.

One-line action: claim the $20 in free credits and run one of your normal tasks through it. Compare the output to whatever you use now.

2. xAI shipped Grok 4.5

On July 8, Elon Musk's xAI released Grok 4.5, a 1.5-trillion-parameter model. Musk called it "an Opus-class model, faster, more token-efficient, and lower cost." It is priced at $2 per million input tokens and $6 per million output, against Claude Opus 4.8's $5 and $25.

If Musk's benchmark comparisons hold up under independent testing, that is not a small gap. That is a top-tier capability at roughly a third of the price. The model also ships inside Cursor (the code editor SpaceX acquired in June) on every plan, including free.

Translation: three labs (Anthropic, OpenAI, xAI) now offer flagship models within a similar quality band. You have real choice, and real price pressure, for the first time.

One-line action: you probably do not need to switch tools. But you might be paying 3x more than you need to for what you actually do.

3. GPT-5.6 finally hit general availability

Remember Sol, Terra and Luna being in government-gated preview last week? On July 9, the whole family moved to broad availability. Priced from $1 (Luna, the cheapest) to $5 (Sol, the flagship) per million input tokens. 1-million-token context window across all three.

Translation: OpenAI is now competing on price alongside quality. The old story of "premium quality costs premium prices" is done.

One-line action: if you were on ChatGPT Plus 6 months ago and stopped renewing because the value felt static, take another look. It is a materially different product now.

4. The quieter story that actually matters most

Anthropic overtook OpenAI on revenue. Anthropic is now running at roughly $47 billion annualised. OpenAI is projected at $25 to $33 billion for 2026.

Translation: it is not just labs shipping models. Paying customers voted, and the leader changed. When enterprise buyers are prepared to spend $47 billion a year on the tool that used to be second place, something structural just shifted.

One-line action: if you have been holding off on paid AI tools because you were waiting to see who wins, the wait is over. There is no single winner, but there is a functional top three now (Anthropic, OpenAI, Google DeepMind), with xAI and Meta coming up fast.

5. The pettier story, for balance

Apple sued OpenAI for trade secret theft. The trigger: OpenAI has now hired more than 400 former Apple employees, and Apple thinks a chunk of them took proprietary knowledge with them on the way out.

Translation: the AI arms race is being fought in HR departments now. This one has nothing you can directly act on. But it tells you exactly how expensive AI talent has become, which flows straight back into what you pay for the tools.

The one honest question

Five stories, one thread.

Which task have you not automated yet because AI was not good enough, or was too expensive?

Because as of this week, neither of those excuses really works anymore. Every single frontier model got materially cheaper and materially better in the last 30 days. If the actual reason you have not moved is that you do not know where to start, that is a solvable problem, and it is not really an AI problem.

The live feed we pull from is here if you want the daily version, it aggregates OpenAI, Anthropic, Google DeepMind, Meta AI, TechCrunch, The Verge, WIRED and MIT into one page:

And if you would rather have ten plain-English AI commands land in your inbox every Monday instead of chasing every headline, the Weekly Command Drop does exactly that. Free.