On January 1, 2026, OpenAI’s updated Rest-of-World Terms of Use took effect. The policy page defines usage rules specifically for users outside the US, UK, and EU — which means most of the world’s markets now share one rulebook, deliberately kept separate from the three biggest ones.
Two other signals belong on the same timeline. OpenAI’s model release notes show the GPT-5.2 base model shipped on December 11, 2025. And on December 9, 2025, CNBC reported that Meta is pursuing a frontier model codenamed Avocado that may be proprietary instead of open source — a marked shift from its Llama strategy. Terms, model cadence, and open-source strategy all converged in the year’s first week.
Tighter Terms: Geography Becomes a Product Boundary
The existence of a Rest-of-World policy is itself the message. When a company carves out “everywhere except the US, UK, and EU” as its own jurisdiction, regulatory divergence has outgrown any single global document. For teams building on OpenAI products, the practical effects land quickly:
- Rights and obligations for users in covered markets now flow from this document, so multi-market products need a compatibility review
- Once terms are regionalized, feature rollouts and data policies can start to diverge by region too
- Legal review now couples to the product roadmap; a terms change has to be managed like a product change
Model Cadence: The Acceleration After GPT-5.2
The GPT-5.2 base model went live on December 11, 2025 — three weeks before the new year. Opening 2026 with the freshest base model already in place usually signals a year of derivative releases and capability expansion on a stable foundation. For API users the trade is clear: shorter effective model lifetimes, migrations becoming routine engineering, and version pinning plus model routing abstraction moving from optional extra to baseline practice.
Two second-order effects follow. Pricing and capability assumptions baked into products last year go stale faster, so unit economics need revisiting more often than annually. And evaluation harnesses gain importance: when models change underneath you, a stable benchmark suite is the only way to tell whether an upgrade actually helps your workload or quietly regresses it.
The Open-Source Split: Avocado’s Turn
CNBC’s report puts a strategic question on the table: Meta’s Avocado may not be open source. If that holds, the Llama-era playbook of trading open weights for ecosystem reach gets a serious revision, and the supply of frontier-grade open models thins out. Teams that depend on open weights should note two things:
- The pool of self-hostable frontier models shrinks, tilting bargaining power toward closed APIs
- Model sourcing needs parallel open and closed paths, so a wrong bet on either side does not strand the product
What the Three Threads Share
Read together, the opening weeks of 2026 point in one direction. Commercial and legal geography is fragmenting, model iteration is running at high frequency, and open source is no longer the industry’s default route. The pragmatic response for developers and product teams is to treat change itself as a design input: assume terms, model versions, and licensing models all shift during the year, and build architectures with room for substitution instead of absorbing each change as an annual surprise.
For Meta, the interesting question cuts the other way. If Avocado ships proprietary, the company is betting that frontier capability is worth more than the ecosystem goodwill open weights generate — a bet the market will score within the year.
Sources
- OpenAI Rest-of-World Terms of Use
- OpenAI model release notes
- Meta pursues frontier model Avocado — CNBC
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.