FABLE 5 — STATUS AS OF JULY 19 EVENING
● Free window closes: Tonight July 19 at 11:59:59 PM PT — the third and final Fable 5 extension
● Opus 5 announced: No
● Fourth extension announced: No
● What starts July 20: Usage credits at $10/$50 per million input/output tokens
● Context: $10/$50 is double Opus 4.8's $5/$25 — the priciest model on Anthropic's price list
● Batch API pricing: Halves both rates to $5/$25 — use batch for any non-realtime Fable 5 work
● Cache hits: $1 per million input tokens — set up prompt caching before spending credits
● Anthropic said: Credit period is not permanent — will restore Fable 5 to subscriptions when compute allows. No date given.
● Other Claude models: Sonnet 5, Opus 4.8, Haiku 4.5 remain fully accessible within plan limits
The End of the Extension Cycle — What Actually Happened
This is the third time Anthropic has pushed back the original July 7 cutoff date — first to July 12, then to July 19. After July 19, Fable 5 moves from included plan access to metered usage credits. The pattern of extensions has been consistent: each one was announced hours before the previous deadline, each was timed to a competitive event, and each maintained the same 50% of weekly limits structure without resetting individual user quotas. Users grew frustrated — in Reddit threads, subscribers said another week of availability offered little benefit if they had already exhausted their included usage.
The Honeycomb EAP signal that appeared in Cursor on July 8 — which pointed to a possible Opus 5 launch before month-end — has not materialised into a public announcement as of tonight. The three scenarios the community identified at the start of the week were: Opus 5 ships (~40%), fourth extension (~35%), credits begin (~25%). Credits begin. Starting July 20, per Anthropic's email to subscribers, all Fable 5 usage runs on prepaid usage credits at $10 per million input tokens and $50 per million output tokens. Anthropic has said it aims to restore the model to subscriptions once capacity allows. No date has been given.
The Credit Economics — What Fable 5 Actually Costs From Monday
The usage-credit rate is $10 per million input tokens and $50 per million output tokens — twice Opus 4.8 and the priciest model Anthropic lists. A heavy 50K-input/15K-output run is about $1.25, so roughly 30 runs a day amounts to $37.50/day or over $1,100 per month. The Batch API halves it to $5/$25, and cache hits cut input to $1 per million. The practical implication: Fable 5 at credit pricing is still the right model for work where its 80.4% SWE-bench Pro accuracy and 1M context window are specifically needed. It is not the right model for general use at $10/$50/M when Claude Sonnet 5 is $2/$10/M (intro through August 31) and Opus 4.8 is $5/$25/M.
| Model |
Input /1M |
Output /1M |
Context |
SWE-bench Pro |
Available in plan |
| Claude Fable 5 |
$10 |
$50 |
1M |
80.4% |
Credits only from July 20 |
| Claude Opus 4.8 |
$5 |
$25 |
200K |
69.2% |
Included in plan ✓ |
| Claude Sonnet 5 |
$2 intro |
$10 intro |
1M |
63.2% |
Included in plan ✓ |
| Grok 4.5 |
$2 |
$6 |
500K |
64.7% |
SuperGrok subscription |
What to Do From Monday
Route general work to Sonnet 5 immediately. Claude Sonnet 5 at $2/$10/M intro (through August 31) handles the vast majority of coding, writing, and analysis tasks. It has a 1M context window — larger than Fable 5's and the same as GPT-5.6 Sol. 63.2% SWE-bench Pro is strong for most production workloads. Default to Sonnet 5 and use Fable 5 credits only for tasks that specifically require 80.4% SWE-bench Pro accuracy or Fable 5's unique architecture advantages.
Set up prompt caching before using Fable 5 credits. Cache hits on Fable 5 cost $1 per million input tokens vs $10 uncached — a 90% input cost reduction. Any workflow with a large system prompt or repeated context should implement caching before spending credits. This is the highest-ROI action you can take before Monday.
Use Batch API for non-realtime Fable 5 work. Batch API halves Fable 5 pricing to $5/$25/M — matching Opus 4.8's standard rate. Any Fable 5 work that does not require real-time output (document analysis, large-scale refactors, batch evaluations) should route through the Batch API from Monday.
Honeycomb/Opus 5 is still possible before July 31. The Honeycomb EAP signal in Cursor has not been officially acknowledged or denied. Anthropic saying credits begin July 20 does not rule out an Opus 5 announcement this week or next. Watch anthropic.com/news and @claudeai on X. If Opus 5 ships with better token efficiency than Fable 5, the credit economics change significantly.
Sources: Anthropic email to subscribers July 12-13, 2026 · ChatForest · Digital Applied · Fable5.app · Vantage Point · Related: Claude Honeycomb EAP — is this Opus 5? → · The $2.49 vs $11.80 per-task cost analysis → · The three scenarios we laid out yesterday →