Claude Fable 5 is available again as of July 1 after a 19 day government ban, and if you use Claude for real work, you now have a decision to make: keep running Opus 4.8, or move your important tasks to Fable 5 while the free access window is open. Three side-by-side tests (same prompts, same account, no head start for either model) give you a clear answer by the end of this post, including the one place where the gap between the two models is impossible to miss.

Why Fable 5 Disappeared, and What Anthropic Changed

Amazon researchers found a way past one of Fable 5’s cybersecurity safety layers, and it was serious enough that the US Commerce Department pulled export approval worldwide, overnight. Anthropic’s fix was a new safety classifier trained specifically to catch that jailbreak. The company reports it now blocks the technique more than 99 percent of the time. That figure is Anthropic’s own number, not an independent audit, so treat it as a claim the company is standing behind rather than a verified fact.

Test 1, The Judgment Call: Both Models Get It Right

The first test is a real business decision with a known right answer: evaluating whether a new sponsorship offer conflicts with four existing deals. Both models reached the correct verdict (no conflict) and neither flagged a false positive. The difference showed up in the depth. Opus 4.8 gave a direct, efficient answer. Fable 5 went further into the analysis, reasoning about what the channel actually does and what the deal means in that context. For a simple judgment call like this, either model works, and honestly a lower tier like Sonnet could handle it too.

Test 2, The Kanban Build: 45 Seconds Faster, Cleaner UI

The second test moves from writing to building: a self-contained HTML kanban board with three columns, card creation, and drag and drop, opened directly in a browser. Both models produced working boards with the same core functionality. Fable 5 finished about 45 seconds sooner and delivered a noticeably nicer user interface. On a simple build like this the two are close, but the speed and polish gap hints at what happens when the task gets harder.

Test 3, The Physics Game: Where the Gap Shows Up

The hardest test asks each model to build a playable game with real physics: using the gravitational pull of planets to slingshot toward a target. This is where the two models stop looking similar. The Opus 4.8 version worked but showed glitches, visual cutoffs, and buggy moments in play. The Fable 5 version delivered dramatically cleaner graphics, gravity that behaves realistically, and a more professional overall experience.

The reason for the difference was visible during the build itself. Fable 5 went back and agentically critiqued its own work, rechecking that what it produced actually functioned before calling it done. Opus 4.8 delivered its output and stopped. That self-checking loop is what Fable 5 is built for, and it is the same capability Anthropic points to for heavier use cases like healthcare analysis and cybersecurity work.

The Free Window Closes July 7

Fable 5 is rolling out globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Pro, Max, Team, and seat-based Enterprise plans can use it free up to 50 percent of their weekly usage limits through July 7. After that, Fable 5 draws from paid usage credits on top of your existing plan. If you want to run your own side-by-side comparison, this week is the week to do it at no extra cost. Heavy builds like the physics game burn real credits, so after the 7th every complex prompt has a price attached.

The Rule: When to Reach for Fable 5

Here is the practical takeaway. For simple, routine tasks (quick judgment calls, comparisons, straightforward writing), Opus 4.8 or even Sonnet gets you there without touching the credit clock. For high-impact, high-visibility work that needs fine-tuning (real code that has to hold up, graphics, motion design, anything where the output gets stress-tested), Fable 5 is the powerhouse, and its habit of rechecking its own work is the difference between code that runs and code that actually works.

The bottom line: both models can do the job, but Fable 5 separates itself the moment the task gets complex. Run your own comparison on a real task from your workflow before July 7 while the free window is open, and let the results decide which model earns the default spot on your desk.