Effort Levels

Effort Levels are Claude Fable 5's five settings that let you buy more reasoning depth by spending more tokens on internal work. The levels Low Balanced Thoughtful High and Max control how many internal steps the model takes before it answers. Low keeps the thinking shallow and the cost down around ten cents for simple SVG tasks. Max opens the throttle so the model can run extended simulations self critique its drafts and explore multiple approaches before committing to output. This is not just marketing. The setting changes the actual compute budget behind the scenes which is why Simon Willison saw his pelican benchmark cost jump from 9.67 cents at Low to 72 cents at Max on day one. For designers this means you can finally tackle those monster tasks like overhauling an entire component library in one shot instead of feeding it one button at a time like you had to with Opus 4.8 last year. The levels tie directly into long horizon capability by giving the model room to maintain coherence across massive contexts and into adaptive thinking by letting it decide when to think harder on tricky subtasks within your design workflows. They also affect token rhythm because higher effort stretches out the invisible work phase before the response appears giving the model time to do better work.

Effort Levels are not temperature settings. They do not make the model more random or creative in the artistic sense most designers chase. They are not a substitute for clear instructions or good system prompts and you cannot replicate the effect by adding think harder at the end of your prompt. They are not free either. Crank everything to Max and you will watch your subscription credits vanish faster than free samples at a design conference. The levels are also not available uniformly across all Claude models with older ones capping out with fewer options and less dramatic differences. They are definitely not something to ignore when working with Fable 5 because the gap between Low and Max can be the difference between a half baked design token file and pixel perfect production ready code complete with documentation and tests.

Take the concrete example of Simon Willison testing his pelican SVG benchmark the day Fable 5 launched on June 9 2026. Low effort gave him a bare bones icon for 9.67 cents that looked like it was drawn by a tired intern at 4pm on Friday. The Medium levels started adding proper structure some of the brand alignment and cleaner code at twenty to forty cents per run. When he hit Max effort the output cost 72.175 cents but delivered an SVG so clean you could drop it straight into a production design system with variants for light and dark mode responsive scaling semantic markup and comments explaining every decision. That single test proved the levels were not hype.

Designers saw identical patterns on actual product work. When Sarah Chen led the migration of a full component library from an old Bootstrap fork to a modern design token system at her fintech company the results varied wildly by level. Low effort fractured the project into seventeen separate chats produced inconsistent spacing scales and introduced three visual regressions that broke the homepage hero. Medium effort handled most components cleanly but still missed contrast ratios on two form elements and required manual review of every generated file. Running the full migration at High effort completed eighty percent of the work in one forty minute session for about eleven dollars and surfaced optimization opportunities in the button hover states that no one on the team had spotted. Max effort pushed completion to one hundred percent caught two contrast bugs suggested three animation easing improvements based on the 2025 brand guidelines and output a complete audit report with implementation notes for the handoff. Total cost thirty four dollars. The same task at Low effort would have eaten three full designer days and still shipped bugs to staging that would have taken another week to fix.

Another team at a SaaS company used Fable 5 for a Figma description to code run on a dense analytics dashboard containing twenty three complex components with intertwined states and conditional logic. The Max effort setting consumed the full one million token context window maintained brand consistency across every button variant hover active and disabled states generated corresponding unit tests for the interactive elements and even proposed two microcopy improvements based on the component labels and existing user research. The Low effort version collapsed after the sixth component started mixing up hover states with active states produced CSS that broke on mobile viewports and ignored half the conditional rendering rules. A third case hit during a design token audit for a client project with two hundred plus tokens. Low effort missed thirty seven inconsistencies and suggested updates that violated the base scale. Max effort caught every mismatch proposed a new modular scale that reduced custom properties by twenty two percent and generated the client presentation deck. These real examples from the June 2026 launch window show that effort levels turn Fable 5 from a fast intern into a senior design engineer when you match the dial to the task complexity instead of leaving it on the default Balanced setting like most people do at first.

Use higher effort levels any time the cost of mistakes is high or the task requires true long horizon capability across massive context windows. Crank it to Max on full design system refactors that touch hundreds of files multi file design token audits that must stay perfectly consistent across light dark and high contrast modes or agentic workflows that stitch together research critique generation and validation over dozens of steps without losing the plot. These are the moments where Fable 5's Mythos class capabilities shine through and where the classifier layer might actually get in the way if your work touches security research or biology. Use Low effort for quick wins like brainstorming icon concepts critiquing a single landing page layout or generating variations on copy text. Do not use Max effort when you are exploring twenty different directions in an hour because you will burn through your June 22 free access window before you finish the first round of feedback and Anthropic starts charging you. Do not use Low effort on anything destined for production code client deliverables or brand touchpoints because the cleanup time will erase any token savings and leave you looking sloppy. Test all five levels against your own patterns now while Fable 5 remains included at no extra charge on Pro Max Team and Enterprise subscriptions through June 22 2026. After that date the wrong choice starts hitting your credit card directly and the difference between levels becomes a real budget line item instead of an interesting experiment.

Effort Levels give designers exact control over how deeply Fable 5 thinks so you stop burning money on tasks that need less and stop accepting weak output on tasks that need more.