The End of the Free Open Model Era

The AI company synonymous with "open and free" has arrived with its first paid API — Meta's announcement of Muse Spark 1.1 is the kind of news that confronts businesses with a fact: the environment for procuring AI models never stops shifting. This learning material reads the announcement as a lesson in "procurement design that does not depend on any single AI vendor," and lays out the preparations Japanese companies in the Philippines should have in place.

According to reports, on July 9, 2026, Meta announced its agentic AI model "Muse Spark 1.1" and began offering the company's first paid developer API (the Meta Model API) in public preview. Pricing is $1.25 per million input tokens and $4.25 per million output tokens, with support for a long context of one million tokens and for tool use and computer use. Availability starts with developers in the United States, with a waitlist for other regions. The pivot by Meta — whose banner had been the free, open-model line — is a prompt to revisit the assumptions behind your AI procurement. Using this case, this material works through vendor-independent procurement design, in order from Part 1 to Part 4.

---

Part 1: Read, Then Draw Out the Implications for Your Company

Three points in the announcement deserve attention.

Step 1: Pre-Reading (3 min)

Before reading, put your own company in the picture.

• How dependent is your company's AI usage on the models of a single provider?
• If that provider's pricing or terms changed, which operations would be affected?
• Have you ever thought through the steps for switching to a different model?

Step 2: First Reading (10 min)

The following is a fictional internal memo, written from the perspective of a Japanese company in the Philippines, based on the facts of the news story.

---

Internal Memo: Meta's Muse Spark 1.1 Announcement and Its Implications for Our AI Procurement

On July 9, 2026, Meta announced its agentic AI model "Muse Spark 1.1" together with the company's first paid API (the Meta Model API). Three points deserve attention.

First, the flagship company of "open and free" has stepped into paid APIs. It is now plain that any plan premised on free models carries an assumption risk: the provider can change course.

Second, the range of choices is nonetheless growing. At $1.25 for input and $4.25 for output (per million tokens), pricing is competitive, and the model is reported to be strong in long context and computer use. More vendors means more negotiating leverage for buyers.

Third, there are regional differences. Availability starts in the United States, with a waitlist elsewhere. For our Philippine operation, there may be a period of "announced, but not yet usable here," so plans premised on the newest model need care.

The implication for our company is a design of "multiple tracks" that avoids over-reliance on any one vendor: (1) decide on one backup model for each major use case, (2) document prompts and work procedures in a form that does not depend on any specific model's quirks, and (3) review pricing and terms quarterly — these three points are what we want in place now, while things are calm.

---

Source: Introducing Muse Spark 1.1 — Meta official blog (July 2026)

Note: The business scenario above is a fictional internal memo created for learning purposes from publicly available facts. Figures and terms of service may change; please refer to the primary source linked above for the latest details.

Step 3: Comprehension Check (5 min)

• What did Meta offer for the first time with this announcement?
• What is the pricing per million tokens for input and output?
• What "regional difference" should a Philippine operation be careful about?

Step 4: The 3-Minute Briefing (10 min)

Practice explaining this story to your management meeting in three minutes. A structure of "what happened (Meta's entry into paid APIs) → why it matters (the free assumption is gone, and choices are increasing) → what we should do (diversify procurement and prepare switching procedures)" makes the briefing land.

Related: Tiered AI Models: Matching Cost to the Job explains this in detail.

Part 2: Key Terms Explained (for Executives)

Open model — An AI model whose weights are published and which you can run on your own servers. It can be used for free, but as this case shows, the terms for new versions can change at the provider's discretion.

Vendor lock-in — The state of being effectively bound to one provider's specifications and prices. With AI, it also arises when your prompts and procedures become over-optimized to one model's quirks.

Public preview — A trial release before general availability. Specifications, pricing, and regions can change at this stage, so incorporate it into production work cautiously.

Computer use — A general term for AI operating a computer screen — clicking and typing — by looking at it. It opens the door to automating routine PC work, but because the blast radius of a wrong click is wide, human supervision must be designed in.

Related: Lessons from the White House 30-Day Pre-Release Review: AI Adoption Planning for an Era When the Newest Models Are Not Instantly Available | Case Study for Japanese Companies in the Philippines explains this in detail.

Part 3: Applying It to Your Own Business

Three things to do in your own business.

Decide a "second choice" for each use case

For each major use case — chatbot, document summarization, coding assistance — choose one alternative to your current model, and run a small check on it once a quarter. Companies that have decided their fallback in peacetime do not panic at price-revision or end-of-service news.

Related: Designing an AI Budget When Costs Halve explains this in detail.

Write prompts and procedures "model-neutral"

Prompts tuned tightly to one model's quirks become liabilities the day you switch. Write instructions generically — purpose, materials, output format — and keep model-specific tweaks in a separate note. That one habit is most of what makes a switch cheap.

Build "not yet available here" into your plans

New models typically launch in the United States first, and the Philippines can be on a waitlist. For plans premised on new features, check availability timing first and pair the plan with an alternative method.

Part 4: Common Failure Patterns (What Not to Do)

Failure 1: Making "because it's free" the core of the business case

Free is the provider's strategy, not a promise. Even when using free models, a one-page memo on "what we do if it goes paid or terms change" changes the quality of the risk you carry.

Failure 2: Switching production right after an announcement

New models look attractive, but in preview, both specifications and pricing move. Verify quality and actual cost in a small test first; switch production only after terms have settled.

Failure 3: Misreading diversification as "contract everything twice"

Diversification is not permanent double contracting. It means having a chosen fallback and occasionally confirming it works. Start with preparations that cost nothing.

Failure 4: Leaving procurement reviews entirely to engineers

Which operations depend on which models is a matter of business risk. The quarterly review should reach management as a one-page list of costs and dependencies.

Tips for Getting the Most Out of This (3 Tips)

1. Draw a one-page "AI vendor map." Your use cases × current models × backup candidates on a single sheet shows you, at a glance, where to look whenever news breaks.
2. Track costs as monthly billed amounts, not unit prices. Token counting differs by model, so reliable comparison is billing-based.
3. Make region-checking part of how you read news. "Announced" does not mean "usable in the Philippines." Make it a habit to check the availability paragraph of a press release first.

Bonus: How to Make Use of PH AI Works

PH AI Works supports Japanese companies in the Philippines, in Japanese, from AI model selection and diversification design to cost visibility and migration testing. If you want to sort out how dependent your company is on which models, feel free to use our free consultation. We can start with a quick assessment of where you stand today.

Sources

• Meta official blog — Introducing Muse Spark 1.1 (July 2026)
• DataCamp — Muse Spark 1.1: Meta's Agentic Model and API