Anthropic redeployed Claude Fable 5 on July 1, 2026, after a three-week suspension. The model went dark on June 12 when US export controls kicked in without warning.
What Happened
June 9: Fable 5 launches alongside Mythos 5 (cybersecurity variant for Glasswing partners).
June 12: US government applies export controls. Anthropic had no way to verify user nationality in real-time, so they pulled both models for everyone, everywhere.
June 26: Mythos 5 access restored for approved US organizations in Project Glasswing.
June 30: Export controls lifted. Howard Lutnick confirmed via X.
July 1: Fable 5 back on Claude Platform, Claude.ai, Claude Code, and Claude Cowork.
The Catch
Pro, Max, Team, and select Enterprise plans get Fable 5 for up to 50% of weekly usage limits through July 7. After that, it burns usage credits. AWS, Google Cloud, and Microsoft Foundry access coming later.
New Safeguards
Anthropic updated their safety classifiers to catch potentially dangerous cybersecurity uses. They also announced an industry jailbreak framework with Amazon, Microsoft, Google, and Glasswing partners — a shared standard for assessing model jailbreak severity.
Why It Matters
Fable 5 is Anthropic’s most capable widely released model. 1M token context. $10/input MTok, $50/output MTok. Adaptive thinking always on. For long-running agents, it’s the new ceiling.
The three-week gap showed how fast a model can disappear when policy shifts. No migration path. No deprecation timeline. Just gone.
Teams building on Fable 5 now have a working model again. But the reminder sticks: frontier model access depends on regulatory weather, not just uptime SLAs.
Sonnet 5 vs Fable 5: Which One for Your Agents?
Here’s the practical comparison for developers building AI agents:
| Aspect | Sonnet 5 | Fable 5 |
|---|---|---|
| Positioning | Cost-performance sweet spot | Flagship capability |
| Best For | 80% daily agentic coding tasks | Complex reasoning, highest-stakes work |
| Pricing | $2/$10 per MTok (intro), then $3/$15 | $10/$50 per MTok |
| Context | 200K tokens | 1M tokens |
| Tool Use | Strong (browser, terminal, API) | Stronger (sustained multi-hour execution) |
| Autonomous Runtime | Medium-length tasks | Multi-hour sessions |
| Availability Risk | Stable (established tier) | Policy-sensitive (just proved by 3-week takedown) |
The takeaway:
If you’re building production agents that run daily — code review, ERP summaries, email triage — Sonnet 5 is your workhorse. It’s got the agentic capabilities for sustained coding and tool use, but at a price that makes sense for recurring workloads. I’ve been running Sonnet 5 for my nightly pipelines and the cost delta vs Fable 5 is 5x for similar output quality on routine tasks.
If you’re building agents that tackle novel, complex problems — architecture decisions, multi-repo refactoring, security audits — Fable 5 is worth the premium. The 1M context window means you can feed it an entire codebase. The adaptive thinking mode handles ambiguity better. But it’s 5x the cost, and the June takedown proved it’s more exposed to regulatory shifts.
My setup: Sonnet 5 for the 80% routine work. Fable 5 reserved for the 20% that needs the ceiling. And a fallback to DeepSeek V4 Pro if either goes dark again.
What’s the backup plan when the model you built on goes dark next time?
Discover more from Susiloharjo
Subscribe to get the latest posts sent to your email.