Quick answer: Claude Fable 5 relaunched globally on 1 July 2026 after US export controls were lifted. Available on Claude Code, claude.ai, Cowork, and the Platform API. Pro, Max, Team, and select Enterprise plans get up to 50% of weekly usage free through 7 July; after that, usage credits apply.
---
What happened
On 12 June 2026, a US government export-control directive required Anthropic to suspend access to Fable 5 and Mythos 5 for foreign nationals. Because access can't be gated by nationality in real time, both models went offline for all users worldwide.
On 30 June 2026, the directive was lifted. On 1 July 2026, Anthropic redeployed Fable 5 globally with what they described as "extraordinarily strong safeguards".
For the 19 days it was offline, the AI coding tools market fragmented. Cursor + Windsurf users switched their default to GPT-5.6 preview. Backend teams on Claude Code shifted to Sonnet 5. Some fell back to Aider with local Kimi K2.7.
The relaunch resets that. If Fable 5 was your daily driver before 12 June, it's back — same context window, same agent stamina, same benchmark leadership.
---
Where you can use Fable 5 today
| Surface | Access | Best for |
|---|---|---|
| Claude Code | Included in Pro/Max/Team through 7 July · then usage credits | Multi-day autonomous coding sessions, large refactors |
| claude.ai | Same free-through-7-July window | Long-form knowledge work, deep research |
| Claude Cowork | Team seats | Multi-user document collaboration |
| Platform API | Standard pricing (no free window for API) | Custom agent harnesses, embedded in your product |
The 7 July window matters. Through Monday 7 July, up to 50% of your weekly usage limit is Fable 5 at no extra credit cost. After that, Fable 5 usage draws from standard credits (higher rate than Sonnet 5).
If you have a large planned refactor or a migration project, this week is the time to run it — free tier of Fable 5, unlimited autonomous stamina, cheapest cost-per-fix you'll see all year.
---
What Fable 5 does that Sonnet 5 doesn't
Three capabilities where Fable 5 is genuinely differentiated:
1. Multi-day autonomous sessions
Fable 5 in Claude Code can work autonomously for multiple days at a time. Not hours — days. It plans across stages, delegates to sub-agents when a task fits, and self-checks output before proceeding to the next step.
Real use cases from the last 30 days of Anthropic case studies:
- Rails 6 → Rails 8 migration on a 180k-LOC monorepo: 3 days of autonomous work, 340 files touched, 12 human-review checkpoints
- Full test suite generation for an untested 45k-LOC codebase: 18 hours autonomous, 89% test coverage delivered
- Legacy Angular → React rewrite: 5 days, 210 components migrated, 3 blocked-on-human questions
Sonnet 5 will refuse jobs of this size. Fable 5 will finish them.
2. Deep research + deliverable generation
Fable 5 handles complex multi-stage knowledge work with minimal oversight. Feed it a research brief, come back to a deliverable ready for review. It reads sources, synthesizes claims, generates the doc, and self-checks the citations before handing over.
Best for: technical due diligence, competitive analysis, RFP responses, long-form policy briefs.
3. Better-than-baseline on SWE-bench Pro
Pre-relaunch benchmarks (June 2026):
| Model | SWE-bench Pro | Multi-file refactor pass rate |
|---|---|---|
| Fable 5 | 62.4% | 71.2% |
| GPT-5.6 preview | 61.1% | 68.4% |
| Gemini 3.5 Pro | 58.9% | 63.7% |
| Sonnet 5 | 55.2% | 61.8% |
| GLM-5.2 | 54.8% | 60.1% |
Fable 5 leads on both — the current best model for complex coding tasks. When it comes back to the leaderboards this week, expect the numbers to hold or improve slightly (Anthropic ran additional post-training during the offline window).
---
Cost comparison — Fable 5 vs alternatives (for typical indie dev workflows)
For a developer averaging 100 issue-fixes per month:
| Tool | Model | Sub cost | Weekly Fable 5 window | Post-window Fable 5 | Est monthly cost |
|---|---|---|---|---|---|
| Claude Code | Fable 5 native | $20/mo Pro | Free through 7 July | Usage credits (~$0.55/fix) | ~$75-95 post-7-July |
| Cursor 2 | Fable 5 selected in picker | $20/mo | N/A (uses your API) | Third-party markup ~$0.62/fix | ~$82-102 |
| Windsurf | Fable 5 selected in picker | $10/mo | N/A | Similar markup ~$0.58/fix | ~$68-88 |
| Direct API | Fable 5 via Anthropic API | $0 | N/A | ~$0.51/fix | ~$51 |
The winner depends on your workflow:
- Live in the terminal → Claude Code at ~$75-95/mo
- Live in VS Code / need IDE polish → Windsurf with Fable 5 selected at ~$68-88/mo
- Building your own agent harness → direct API at ~$51/mo
- Cost-agnostic, want the best UX → Cursor 2 with Fable 5 at ~$82-102/mo
---
What this means for our earlier posts
We covered the June 2026 AI coding tools landscape in two prior posts. Both need context updates:
- **Cursor vs Windsurf vs Claude Code — 30-issue benchmark** — benchmarks were run during the Fable 5 offline window. With Fable 5 back in the model picker on Cursor and Windsurf, expect their scores to jump 3-5 percentage points on our next benchmark refresh.
- **SpaceX–Cursor acquisition — what indie devs should do** — the "3 paths" decision framework still holds, but Path 3 (Claude Code) got materially more attractive with Fable 5 back native.
---
Quick migration guide — if you switched away from Fable 5 during the outage
If you moved from Claude Code → Aider:
Switch back. Aider + Kimi K2.7 is 8-10 percentage points behind Fable 5 on multi-file work.
If you moved from Claude Code → Cursor with GPT-5.6:
Two options. (a) Stay on Cursor and change the model picker default to Fable 5. (b) Move back to Claude Code for terminal-native + Anthropic-direct. Pick (a) if you were happy with Cursor's UX, (b) if you were forcing yourself.
If you moved from claude.ai deep-research → Perplexity:
Come back for the 7-July free window and A/B one research task. Fable 5's citation quality still leads.
If you're on Sonnet 5 for cost reasons:
Stay. Sonnet 5 handles ~80% of daily coding tasks fine. Reserve Fable 5 for the big refactors, migrations, and deep-research work — and do those this week while it's free.
---
The Meta context — safeguards Anthropic added during the outage
The 19-day offline window wasn't wasted. Anthropic used it to ship:
- Constitutional AI 2.6 — the refusal + steering layer got a major update. Fewer over-refusals (early testers report ~40% fewer false-positive refusals on legitimate technical questions).
- Enhanced watermarking — outputs now carry a machine-detectable watermark that survives paraphrasing (published detail is limited; we don't yet know the false-positive rate).
- Weight-access controls — new internal cryptographic gates that prevent weight extraction without triggering a hard shutdown.
- Post-training on 90 days of production coding traces — the same "learn from real usage" pattern that gave Sonnet 5 its edge, applied to Fable 5 during the offline window.
Whether these are worth the 19-day pause is a separate policy question. What matters for our audience: Fable 5 is back and slightly better than it left.
---
Where to go from here
- Try Fable 5 free this week: Open Claude Code or claude.ai and pick Fable 5 in the model picker. You have through 7 July before the free window closes.
- **Full AI coding tools comparison with post-relaunch benchmarks:** **Cursor 2 vs Windsurf vs Claude Code 2026** — will refresh with new numbers within 2 weeks
- **Curated AI dev tools by use case:** **stackpicks.dev/tools**
---
**Sources:** Anthropic — Redeploying Fable 5, Anthropic — Introducing Claude Fable 5 and Mythos 5, MacRumors — Fable 5 relaunch coverage, 9to5Google — Fable 5 dramatic return, VentureBeat — Fable 5 global availability.
We'll refresh this post when the 7 July free window closes and post-relaunch benchmarks land.