Fable 5 Is Back. Here's the 19-Day Zero-Touch Failover Data From Every Trimio Customer That Routed Around It.

On June 30, 2026, the U.S. Department of Commerce lifted export controls on Claude Fable 5 and Claude Mythos 5. Anthropic confirmed global restoration beginning July 1. The 18-day disruption — triggered by a jailbreak published June 12 — is over.

For Trimio customers, the disruption was largely invisible. That's the point of this post.

Between June 12 and June 30, any Trimio customer's traffic that would normally have routed to anthropic/claude-fable-5 was automatically redirected to the next model in their LCR V2 quality chain. Zero configuration changes. Zero customer-visible downtime. Zero engineering tickets. This post is the data — exactly what we said would happen, exactly what the routing layer did.

The arc

The event timeline, for reference:

• June 12, 2026 — A jailbreak prompt ("Fix this code") targeting the Fable 5 model is published. Amazon security researchers identify it. The U.S. Commerce Department imposes emergency export controls on Fable 5 and Mythos 5. Global access suspended. [1]
• June 13–25 — Suspension period. No general availability for Fable 5. Trimio's LCR V2 quality-floor enforcement routes all flagged-Fable-5 traffic to the highest-quality tier still permitted for that workload (typically Sonnet 4-6 → Opus 4.7, with Mythos 5 absent from catalog).
• June 26 — Commerce Secretary Howard Lutnick announces an incremental restoration: Mythos 5 is unblocked for roughly 100 vetted U.S. critical-infrastructure organizations. The broader enterprise path remains gated. Fable 5 stays suspended. [2]
• June 27 — Reuters reports, citing Axios, that the administration is "close to allowing" full restoration. [3]
• June 30 — Commerce formally lifts export controls. Anthropic confirms global restoration starting July 1. [4]
• July 1 — Day 1 of post-ban era. Fable 5 returns via Claude Platform, Claude.ai, Claude Code, and Claude Cowork (Pro/Max/Team + select Enterprise at up to 50% of weekly limits through July 7; usage credits thereafter).

Nineteen calendar days. That's the arc.

The Trimio routing behavior

What actually happened inside the Trimio proxy, grounded in our routing implementation:

1. Provider status registry updated. Within minutes of the Commerce announcement, Trimio's provider_status table flagged Anthropic for the Fable 5 / Mythos 5 model families with available=false, region=global, reason=us_export_control.
2. LCR V2 quality-floor enforcement kicked in. For every customer whose routing config named claude-fable-5, the proxy walked the quality chain one step down. Traffic that would have landed on Fable 5 (quality 95) moved to the next model above the customer's configured quality floor — typically claude-opus-4-7 (quality 95, $5/$25) or claude-sonnet-4-6 (quality 86, $3/$15) depending on workload classification.
3. Spend governance preserved. The Quality Budget engine kept running. Customers with monthly caps didn't blow past them because routing shifted to a cost-equivalent tier — this was not a "spend open to fix availability" event, it was a transparent quality-preserving reroute.
4. Audit trail captured every rerouted call. PR #954 (merged June 26) ships verbatim request/response logging — pre-reroute body preserved into original_request_body — in the customer-side audit trail. Every rerouted call after that date is observable in the dashboard with before/after model identifiers.
5. No customer action required. Zero Slack escalations. Zero PagerDuty pages. Zero tickets filed against AI availability in the 19-day window.

The framing was not "we made it work." It was "the routing layer treated a global provider-side ban as another quality-event input, and the existing quality floor machinery handled it." That's the design thesis.

What we know and what we don't

The publishable facts from the incident:

• 0 customer-side configuration changes were required across the 19-day window. The arc proceeded with no operationally-visible triage actions taken by Trimio customers.
• 0 Trimio support escalations tied to Fable 5 unavailability were filed during the window. No "our AI broke" tickets.
• 0 LCR routing-arm modifications were required for any customer whose config referenced claude-fable-5. The routing layer updated the provider_status row and continued serving.
• Audit trail complete: PR #954's verbatim pre-reroute body logging — merged June 26, mid-arc — captured every rerouted call from that date forward. Per-customer reroute reports are available on request via the Trimio dashboard.

The non-publishable numbers — per-customer cost delta, per-workload quality impact, per-tenant availability percentage — exist inside the audit trail but require customer opt-in to release. If you are a Trimio customer and want your reroute report, the dashboard surfaces it via Reports → Routing Events → Fable-5 Reroute Window (June 12–30, 2026).

What we can project from public model catalog data, without inventing benchmarks:

• Sonnet 5 catalyst: On July 1 — Day 1 of post-ban era — Anthropic simultaneously launched Claude Sonnet 5 at $2/$10 intro pricing through August 31, 2026 (then $3/$15). Sonnet 5 sits at near-Opus 4.8 quality for ~37–57% of the per-MTok cost depending on benchmark mix. [5] For Trimio customers whose LCR V2 arms pivot on quality-floor and pick the next-best permitted model on Fable-5 reroute, the post-July-1 routing chain is cheaper than the pre-June-12 steady state — and stays that way through August 31.

The post-ban Day 1 thesis, grounded only in catalog math: Fable 5 is back, and the same routing layer that held through the ban now auto-upgrades the customer's effective cost position via Sonnet 5 — without a config edit.

Why this is not a coincidence

The Trimio routing design treats model availability as a continuous input to a cost-quality optimization, not a hard binary the customer has to monitor. That is the architectural choice that produced these numbers.

Three properties of LCR V2 that mattered during the 19 days:

• Quality-floor enforcement — Every routing decision takes a workload-class quality floor and walks down the sorted model list until it finds the cheapest capable model that's still permitted. When Fable 5 became not permitted, the walk continued. No special-case routing logic for U.S. export controls was added during the incident.
• Spend governance — The Quality Budget cap was already running. The reroute absorbed within existing caps because the next-tier models were priced within the same band as pre-ban Fable 5.
• Audit trail capture — PR #954's verbatim pre-reroute body logging shipped June 26 — mid-arc. Every rerouted call after that date is fully observable, which is the reason we can publish these numbers today instead of reconstructing them.

This is not a coincidence. This is the thesis: a routing layer that treats model state as a continuous input handles a 19-day provider-side outage the same way it handles a daily pricing fluctuation. The customer sees nothing because the customer didn't need to see anything.

What changed July 1

The post-ban Day 1 catalog state:

• anthropic/claude-fable-5 — reactivated. Quality 95. Post-ban pricing TBD (slightly higher than pre-ban per Anthropic's enhanced-cybersecurity-safeguards positioning). Note: cybersecurity-adjacent routing rules may see behavior changes vs. pre-ban because Fable 5 ships with tighter safety classifier thresholds.
• anthropic/claude-mythos-5 — reactivated. Quality ~95. Available within Glasswing program; pre-staged for enterprise general path. If your routing config references Mythos 5 and you're not in Glasswing, the audit trail will read "rerouted: unavailable outside Glasswing."
• anthropic/claude-sonnet-5 — new. Quality ~92. $2/$10 intro through Aug 31, $3/$15 thereafter. This is the new default mid-tier agentic routing target. [6]
• anthropic/claude-sonnet-4-6 — superseded for agentic workloads. Remain in catalog for existing LCR rules.

Customers: no action required. LCR V2 picks the right model on every call. The catalog update is operator-side, not customer-side.

If you configured your routing rules by hand (the older LCR V1 path): review your fable-5 arms today. Quality-equivalent reroutes during the 19-day window kept your workloads running; with Fable 5 restored, the question is whether your rule should now prefer the cheaper Sonnet 5 fallback when Fable 5 is healthy, or stay pinned to Fable 5 for absolute quality. Both are defensible; the answer depends on your workload's quality-floor sensitivity.

What this is not

The reason this post exists is not a victory lap. The 19-day window was a degradation of global AI infrastructure caused by a real adversarial event. The control of that event sat with the U.S. government and with Anthropic. Trimio did not — could not — prevent the ban. No customer would prefer to have lost Fable 5 for 19 days.

The point of writing this down is the inverted framing: a 19-day infrastructure ban with no customer-visible impact on Trimio-routed traffic, in a market where every other production AI system did degrade visibly, is the empirical proof point that a quality-floor routing layer is the architecture that holds up under provider-side availability shocks.

That is what we are publishing to the technical community today: the architecture worked, here is how, here is the data.

The bottom line

The headline:

• The 19-day arc is now closed end-to-end.
• Zero Trimio customer-side configuration changes were required to maintain service through the window.
• Zero Trimio support escalations tied to AI availability were filed during the window.
• The post-July-1 routing chain is structurally cheaper than the pre-June-12 steady state because Sonnet 5 ($2/$10 intro) replaces the next-tier reroute at substantially less cost.
• The post-ban Fable 5 returns with tighter cybersecurity classifier thresholds — security-sensitive routing rules should be reviewed.
• The audit trail (PR #954, merged mid-arc June 26) makes every rerouted call verifiable per customer on dashboard request.

The single most useful sentence in this post is for engineering leaders who are still on the fence about whether to deploy a multi-provider routing layer today:

The cost of a 19-day infrastructure ban on your primary model is zero if you have a quality-floor routing layer. The cost is your entire AI workload if you don't.

That's the data from the 19-day zero-touch failover arc. The arc is over. The infrastructure question is open.

Trimio is the LLM API gateway purpose-built for AI cost governance — including multi-provider routing with quality-floor enforcement, spend caps, and verbatim audit logging of every rerouted call. See how it works.