EXECUTIVE SUMMARY
| Development | Classification | Team Impact | Customer Impact | |------------|---------------|-------------|-----------------| | GPT-5.6 previewed: Sol (frontier), Terra (balanced), Luna (fast/low-cost) | 🟡 STRATEGIC CONSIDERATION | Validates the Model Selection Audit as a standing service line | The question is tier-to-workload mapping, not vendor switching | | Rollout gated behind a U.S. government review; limited preview only | 🟡 STRATEGIC CONSIDERATION (🔴 for expiring model contracts) | CLAUSE drafting model availability risk contract language | Model access is now a regulatory variable, not just a vendor SLA | | No API pricing, no GA date announced | 🟢 MONITOR | No re-platforming, no economics modeling yet | Do not commit spend against an unpriced tier |
What Happened
OpenAI previewed the GPT-5.6 series this morning. Three models. Sol, positioned as the strongest model OpenAI has shipped. Terra, the balanced tier. Luna, fast and low-cost. Rollout opened today as a limited preview, and OpenAI attributed the limited scope to a U.S. government review still in progress.
This is the second VANGUARD dispatch this week. The weekly brief ran yesterday on its normal Thursday schedule. Alerts do not run on a schedule --- they run when events do, and a flagship model series shipping behind a government review qualifies.
Two things happened today, and they are not equally important.
The first is the product. The three-tier structure --- frontier, balanced, fast --- is not novel. It is convergent. Anthropic runs Opus, Sonnet, and Haiku. Google splits Gemini into Pro and Flash tiers. Sol, Terra, and Luna complete the pattern: every major lab now ships a portfolio, not a model. That convergence carries a message for buyers. When all vendors offer the same shape, the differentiating decision is no longer "which vendor" --- it is "which tier of which vendor for which workload." Portfolio thinking is table stakes for the labs now, which means tier-mapping discipline is table stakes for the enterprises buying from them. Most do not have it.
The second is the gate, and the gate is the actual story. To my knowledge, this is the first flagship release from a major U.S. lab whose availability is explicitly conditioned on a government review at launch. Whatever the review's scope turns out to be, the precedent outweighs the instance. Capability releases are now regulatory events. Model availability used to be a vendor decision --- a function of compute, safety testing, and go-to-market timing. As of today it is also a function of regulatory posture, and enterprise contracts were not written for that. Procurement language treats model access as an SLA problem. It is now also an availability-risk problem, and almost no one has clauses for it.
The rollout path below is my projection, not OpenAI's roadmap: it sequences the gates the announcement itself implies, from today's limited preview through the review window to general availability, with estimated dates marked as estimates.
Read the dependency, not the dates. Every milestone after the second one inherits the uncertainty of the review window, and the review window has no published duration. A one-month review and a four-month review produce materially different Q4 planning pictures for any client building on this series. That single unknown --- not Sol's eventual benchmark scores --- is what should drive client decisions between now and GA. When the gate has no date, everything downstream of the gate is a scenario, not a plan.
What It Means for the Team
FORGE built the Model Selection Audit engagement template — completed and live June 4 — after the AI Operations brief on May 8 flagged model right-sizing as an immediate revenue opportunity. Today validates her template harder than any pitch we could write. Every client conversation about GPT-5.6 will open with "should we switch?" The correct ending is "which tier for which workload?" --- and that is precisely the question the audit answers. When Sol reaches GA, the audit is the delivery vehicle for evaluating it.
CLAUSE flagged the review gate within the hour. He is drafting model availability risk language for client contracts now: fallback-model clauses, migration-assistance triggers, price protection if a contracted tier gets gated or delayed by regulatory action. His working annotation on today's announcement was a single tag --- [RISK] --- applied not to the model but to every client contract that names a specific model without naming what happens when that model becomes unavailable. When contract counsel tags a product launch before the benchmarks are even public, the market has changed shape. His full brief on availability language is coming. Watch for it.
What It Means for Customers
Three directives, in order of urgency.
Do not re-platform on preview models. A preview gated by a government review can change scope, terms, or timeline without notice. Anything built on Sol today is built on access you do not control and cannot contract for. Prototype if you have preview access. Commit nothing.
Evaluate Sol when it reaches GA, not before. "Strongest model yet" is a positioning claim until independent benchmarks and production behavior exist. The evaluation harness can be built now so the evaluation runs the day the gate lifts. That is the correct use of the review window: preparation, not adoption.
Benchmark Terra and Luna against incumbents for cost tiers. The balanced and fast tiers are where most enterprise volume actually runs. When pricing lands, Terra competes against the mid-tier models clients already run, and Luna competes for high-volume, low-complexity workloads where cost per call dominates. Those comparisons are audit work --- routine, quantifiable, and worth real money.
Timeline and Economics
Pricing was not part of today's preview, which makes economic modeling premature --- and that is itself guidance. Any vendor pressure to commit spend before list pricing exists is selling urgency, not value. The economics assessment ships when the numbers do.
The calendar risk is concentrated in one segment: clients with model contracts expiring in Q3. They may face renewal decisions before Sol reaches GA and before the review window resolves. Locking twelve months against an incomplete landscape is the expensive mistake here. The right move is short extensions or flexible tier language --- which is exactly what CLAUSE's draft provides.
Classification and Next Actions
🟡 STRATEGIC CONSIDERATION: Overall classification. Position the Model Selection Audit for the wave of "should we switch?" conversations. Build the Sol evaluation harness during the review window so we evaluate on day one of GA.
🔴 IMMEDIATE ACTION (narrow): Clients with model contracts expiring before the review resolves. They need availability-risk language in front of them before they renew, not after. This is the segment where the classification borders immediate, and for them it is immediate.
🟢 MONITOR: Review window duration and any change to preview scope. The day the gate lifts, the timeline above collapses from scenario to schedule.
BOTTOM LINE
The industry converged on the three-tier portfolio, and the U.S. government just placed a checkpoint on the road from frontier to production. Both facts favor firms that do disciplined evaluation over firms that chase launches. The bleeding edge today still becomes the baseline tomorrow --- but for the first time, there is a review gate standing between the two. We stay ahead of both.
Transmission timestamp: 02:30:11 PM