The situation in plain terms: Claude Fable 5 is no longer free. The 14-day free evaluation window that opened on 9 June has closed. From 23 June, using Fable 5 requires usage credits. Claude Sonnet 4.6 remains fully available on your plan with no credits required. The decision in front of you is straightforward — but it is worth making deliberately rather than by default.
What happened: the timeline
Anthropic launched Claude Fable 5 on 9 June 2026 with a 14-day free access window on all paid plans. The model represented the most capable publicly available release from Anthropic at the time of launch — with improvements in reasoning, long-document analysis, instruction-following, and multimodal tasks compared to previous Claude generations.
On 23 June, as announced at launch, usage credits became required to access Fable 5. The credit structure is $10 per million input tokens and $50 per million output tokens, purchased through the Anthropic console. Claude Sonnet 4.6 — the standard workhorse model — remains plan-included with no usage credit requirement.
Usage credits are purchased in advance through the Anthropic console. A $10 credits purchase gives you 1 million input tokens on Fable 5 — roughly equivalent to 750,000 words of input, or approximately 1,500 moderately-sized documents processed. For most UK SMBs using Claude for drafting, analysis, or customer communication, $10 of credits could represent several weeks of Fable 5 usage at a reasonable workload. The cost is not prohibitive for high-value tasks — the question is whether the quality improvement over plan-included Sonnet 4.6 justifies the additional spend for your specific use cases.
The API disruption context
There is additional context worth noting. A US export control directive issued on 12 June 2026 affected Fable 5 API access, taking the model offline at the API level for several days. Access was restored on 18 June with new identity verification requirements for certain jurisdictions. UK users were not directly subject to the new restrictions — the export control measures were targeted at specific jurisdictions outside the UK — but the disruption affected API availability and may have cut into the usable evaluation window for some UK users who access Claude via the API rather than directly through the console.
If you were attempting to evaluate Fable 5 during the 12–18 June window via the API and experienced access issues, this is why. The model is now fully accessible again for UK users.
What you do not need to do
You do not need to panic, switch providers, or immediately purchase credits. Claude Fable 5 is still there and still available — it simply costs money to use now. Sonnet 4.6 is still excellent and plan-included. The model market is not going anywhere. This is a pricing transition, not a service disruption.
You also do not need to treat this as a permanent decision. Usage credits are purchased in small increments. You can buy $10 or $20 of credits, run a structured comparison on your most demanding workflow, measure the output quality difference, and then decide whether to continue. If the improvement is marginal for your use case, you stop buying credits. If it is material, you build the cost into your AI budget as a line item.
How to make the decision
The one decision: run a real comparison, then choose deliberately
The only way to make this decision well is to compare Fable 5 and Sonnet 4.6 on your actual highest-stakes workflow — not a toy test, not a general benchmark. Choose the task where AI output quality has the most direct consequence for your business: a client-facing proposal, a complex analysis, a sensitive customer communication, a legal or compliance document. Run the same task on both models using identical inputs. Evaluate the output honestly against your quality criteria. If Fable 5 delivers a material improvement on that task, estimate your monthly token volume and calculate the credit cost. A useful rule of thumb: if the improvement is worth more than the monthly credit cost in time saved or quality improvement, buy credits. If it is not, use Sonnet 4.6. This comparison takes 30–60 minutes and is worth doing before you either commit to a credits budget or permanently write off the model.
Sonnet 4.6 is likely sufficient for: everyday drafting, customer email responses, standard document summarisation, routine data extraction, social media content, FAQ responses, straightforward research tasks. Fable 5 may justify credits for: complex multi-step reasoning tasks, long-document analysis (50+ pages), high-stakes client-facing output where quality directly affects commercial outcomes, technical coding or architecture tasks, nuanced tone-matching for senior client communications. The more consequential and intellectually demanding the task, the more likely Fable 5's improvements will be perceptible and valuable.
The broader lesson for AI tool management
The Fable 5 pricing transition is a preview of how AI model management is likely to work going forward. Free evaluation windows followed by usage-based pricing is becoming a common pattern. The businesses that benefit most are those with a lightweight evaluation habit already in place — who can test a new model quickly when a window opens, reach a grounded decision within days, and implement accordingly. The businesses that lose out are those who defer evaluation until after the window closes and then make decisions based on general reputation rather than evidence.
Building a 30-minute AI evaluation protocol — same task, same input, same quality criteria, model A vs model B — is worth doing once and keeping as a standing process. You will need it repeatedly as the model market continues to move at pace.