June 9: launched. June 12: pulled by the US government. July 1: back. July 7, 12, 19: three deadline extensions. July 20: permanently settled. Six weeks that changed how every business thinks about AI model risk.
If you've been following the Fable 5 story and lost track of where things stand — this is the complete update. If you missed it entirely — here's the full timeline, and why it matters for how you build with AI tools going forward.
🔧 What Actually Happened — The Complete Timeline
June 9, 2026: Anthropic launches Claude Fable 5 — their most powerful publicly available model ever, and the first time a Mythos-class model was accessible to general users. Also launched: Claude Mythos 5, a version with fewer safety constraints, limited to a small set of vetted organisations in Project Glasswing.
June 12, 2026 (5:21pm ET — three days after launch): Anthropic receives a letter from US Commerce Secretary Howard Lutnick invoking export control law. The directive: suspend access to both Fable 5 and Mythos 5 for any foreign national, anywhere in the world, effective immediately. The trigger: Amazon researchers had found a method of bypassing Fable 5's safeguards that allowed it to identify software vulnerabilities in a codebase. Anthropic's response — faced with no reliable way to verify user nationality in real time — was to shut both models down completely for every single customer, regardless of nationality. All other Claude models remained unaffected.
June 12–30 (19 days): The suspension runs through three weeks of negotiations, government review, and public disagreement from Anthropic. The company argued the flagged technique wasn't unique to Fable 5 and was already present in Claude Opus 4.8, GPT-5.5, and other freely available models. They were right — the government ultimately accepted this. Meanwhile, businesses that had built workflows around Fable 5 scrambled. Those with fallback models set up survived the disruption in under an hour. Those without spent days rebuilding under pressure.
June 30: US Commerce Department lifts the export controls. Anthropic restores Fable 5 globally the following day.
July 1: Fable 5 is back. Available on Claude.ai, Claude Code, Claude Cowork, AWS Bedrock, Google Cloud, and Microsoft Foundry. A new safety classifier now sits in front of the model, blocking the specific jailbreak technique in over 99% of attempts. When it blocks a request, it automatically routes to Opus 4.8. Included in paid plans at up to 50% of weekly usage limits — initially through July 7.
July 7 (hours before the deadline): First extension. Anthropic pushes the free access window to July 12. No other terms change.
July 12 (again, hours before the deadline): Second extension, to July 19. Also: Claude Code's weekly rate limits stay 50% higher through the same date. Community frustration with the rolling cliffs is real — Polymarket had been pricing another extension at 25% probability before each one landed.
July 13: Anthropic extends Fable 5 access a third time — now through July 19 — timed visibly as a competitive response to OpenAI's GPT-5.6 launch the same week.
July 18: Anthropic announces the permanent resolution.
July 20 (today): Fable 5 is now a standing feature of Max and Team Premium plans at 50% of weekly usage limits. No expiry date. No more cliffs. Pro and Team Standard users get a one-time $100 usage credit, then pay API rates ($10 input / $50 output per million tokens). Claude Code weekly limits stay 50% higher through August 19.
The verdict: The chaos is over. The lesson it leaves behind is worth keeping.
🧪 Real Business Example
During the 19-day suspension, a UK-based software agency that had built their code review workflow around Fable 5 via the API faced an immediate operational problem — errors from the night of June 12, no warning, no timeline for return.
Teams that had Opus 4.8 already configured as a fallback changed one line — the model string — and were running again within an hour. Teams that hadn't spent several days rebuilding workflows on the fly.
Same disruption. Completely different experience based on one decision made before June 12.
Post-restoration, the same agency ran a quiet benchmark: Fable 5 versus GPT-5.6 Sol on their core code review tasks. The result matched what Artificial Analysis published independently on July 18 — Fable 5 scores approximately 59.9 on their composite intelligence index, Sol approximately 58.9. Effectively a tie on general capability. But Sol costs $5 input / $30 output per million tokens versus Fable 5's $10 / $50 — and completes equivalent tasks at roughly one-third the per-task cost because it emits fewer output tokens.
The agency is now running Fable 5 on Max plans for complex work, and Sol via the OpenAI API for high-volume tasks where cost per run matters. Both fallbacks. Both planned.
📋 Step-by-Step: What to Do Based on Your Current Plan
If you're on Max or Team Premium:
Fable 5 is now permanently included at 50% of your weekly usage limits — no action needed, no credits to enable
It draws down weekly limits faster than Opus 4.8 or Sonnet 5 for equivalent work — reserve it for tasks where frontier reasoning matters, not routine drafts
When Fable 5 is blocked by the new classifier (rare, under 5% of sessions), you'll be notified and routed to Opus 4.8 automatically
Claude Code's weekly limits are 50% higher through August 19 — if you use Claude Code, this is extra runway to use before that resets
If you're on Pro or Team Standard:
You received or will receive a one-time $100 usage credit — enough for roughly 2 million input tokens of Fable 5 access
Use that credit window for your most complex, highest-stakes work — board reports, strategic analysis, long document review, anything where frontier reasoning genuinely changes the output
After the credit runs out, Fable 5 is available at $10/$50 API rates — or consider upgrading to Max if you find yourself needing it regularly
Sonnet 5 remains on introductory pricing ($2/$10 per million tokens) through August 31 — the most cost-efficient model in the lineup for most everyday work
For everyone: 5. Keep your fallback documented and ready — the Fable 5 situation resolved, but the export control framework that triggered it is still in place. A second incident, with any frontier model, remains possible on the same timeline: immediate, without warning, legally binding. 6. Don't dismantle whatever backup workflow you built during the 19 days — if Opus 4.8 or GPT-5.6 handled your work during the suspension, keep those integrations live.
❓ The Dumb Question
"Anthropic extended the deadline three times in two weeks. Does that mean the $100 credit for Pro users is a bad deal?"
Honestly — yes, for heavy users. The three extensions meant that Pro subscribers who were using Fable 5 daily got six weeks of included access before the credit switch. Going forward, Pro gets a $100 one-time credit (roughly 2 million input tokens) and then API billing at $10/$50 per million. At that rate, a heavy user doing several long Fable 5 sessions per week would burn through the credit in under a fortnight and face real ongoing costs. The math only works if you use Fable 5 selectively — for complex, high-value tasks where its edge over Opus 4.8 actually justifies the cost. For everything else, Sonnet 5 at $2/$10 is the right call until August 31 introductory pricing ends. The Max plan at $100/month makes more sense than Pro if Fable 5 is a regular part of your workflow.
💰 Where Things Stand Now — Clean Summary
Plan | Fable 5 Access | Cost |
|---|---|---|
Max (5x or 20x) | Permanent, 50% of weekly limits | $100–$200/month (plan cost) |
Team Premium | Permanent, 50% of weekly limits | Team pricing |
Pro | One-time $100 credit, then API rates | $20/month + credits |
Team Standard | One-time $100 credit, then API rates | Team pricing + credits |
API (direct) | Full access, no cap | $10/$50 per million tokens |
Fable 5 vs GPT-5.6 Sol (benchmarks) | ~59.9 vs ~58.9 — effectively tied | Fable 3x more expensive per task |
One honest note on the benchmark comparison: Fable 5 holds a narrow lead on repository-level software engineering tasks. Sol leads on terminal-based and agentic coding. For most small business owners doing writing, analysis, and planning tasks — the difference is not meaningful. The cost difference is.
⚡ The Practical Play
This week: if you're on Max or Team Premium, open a Fable 5 conversation and test it on the most complex document or analysis task in your current workflow. Compare the output quality to what Opus 4.8 or Sonnet 5 produces. That comparison tells you whether Fable 5 earns its place in your workflow — or whether you've been given a premium model you don't actually need for your specific use case.
📰 News That Matters
Two developments worth watching as of this week. First: a model called "Honeycomb" was discovered inside Cursor's editor by developers running tests — and Cursor was acquired by SpaceX in a deal that included access to Anthropic's model pipeline. Honeycomb is widely speculated to be an early version of Opus 5. Anthropic hasn't confirmed anything. Second: Anthropic confidentially filed for an IPO just 11 days before the Fable 5 suspension hit. The export control episode added regulatory risk to the IPO story in a way that wasn't there before. How Anthropic handles model access and government relations going forward is now a public markets question, not just a product one — which means the incentives around model reliability and uptime are, if anything, higher than they were before June 12.
🚫 Skip This
Treating the July 20 permanent inclusion announcement as a sign that Anthropic's model access is now stable and predictable. The export control framework is unchanged. A new frontier model launch that triggers government review could produce an identical disruption on an identical timeline — immediate, global, no warning. The right response to "it's settled" is not to relax the fallback plan. It's to build redundancy into your AI infrastructure the same way you'd build it into anything else that's business-critical. The businesses that learned that lesson during the 19-day suspension didn't unlearn it when Fable 5 came back. Neither should you.
Until next issue, Kris
The Layman's AI — The only AI updates your business actually needs.