Theo reports Fable 5.2 uses $200 Ultra allowances faster than Opus

• The $200 Ultra plan yields about $3,021 of Fable 5.2 usage versus $14,662 of Opus 5.5 usage at API list prices, according to theo's estimates.
• Fable is capped at half the weekly allowance in his usage notes; exhausting Fable and then using Opus brings the monthly projection to $10,353.
• Completed threads look closer than the dollar figures: his weekly projection estimates roughly 900 mixed-model turns versus 1,200 Opus-only turns, but slightly more threads.

Anthropic's optional fast mode doubles Fable's API token rates to $16 input and $80 output per million tokens. Its launch reset rules restore weekly usage spent on Fable while leaving other-model usage unchanged.

Theo values projected subscription traffic at API list prices. Ultra's pricing page separately includes $200 a month in actual API credits.

Theo based the calculations on more than 1.1 million logged requests. The chart's archive spans August 11 through November 9, 2026, and its monthly estimates extrapolate 30 days from approximately 4.3 weekly allowances.

He said in a reply that data from himself and Julius was sufficient to validate most of what he wanted to know. Before Fable's release, he had reported running 8 to 25 Opus 5.5 threads for much of a day while using less than 20% of a $200 subscription.

Theo himself called API prices “insanely inflated”. He also corrected the original post's model name from “Fable 5.5” to Fable 5.2.

Anthropic's plan documentation describes overlapping limits:

• All models draw from the regular weekly allowance.
• Fable models collectively can consume up to 50% of that allowance and draw it down faster than other models.
• After reaching the Fable cap, further Fable usage requires usage credits; another model can consume the remaining subscription allowance.

Within Theo's monthly projection, equal halves of the quota yield $3,021 of Fable or $7,331 of Opus. That implies roughly 2.4 times as much quota consumed per API-equivalent dollar for Fable, before accounting for the restriction to half the weekly pool.

Fable's tendency to finish work in one prompt is the interesting wrinkle in Theo's weekly comparison:

• Opus only: approximately 1,200 turns across 230 threads.
• Fable until its cap, then Opus: approximately 900 turns, but slightly more threads.

The mixed projection has about 25% fewer turns. Theo cautioned that much of his Fable usage had been benchmarking the new model, so the workload mix could change.

Per-million-token prices fell 20%, as Theo's pricing comparison and Anthropic's API pricing show:

• Input: $10 for Fable 5.1 → $8 for Fable 5.2.
• Output: $50 → $40.

Anthropic's FrontierCode 2.0 results compare specific effort settings: Fable 5.2 at High scored 45.3%, approximately matching Opus 5.5 at Max's 45.5%, for roughly half the API cost per task. Those benchmark dollars and Theo's subscription-quota estimates measure different things.

Theo priced that run at approximately $42 in tokens and judged it one of his better Fishslop builds, though not significantly better than Opus 5.5 given the price.

Another early tester, zeeg, ran a multi-day 3D game task with one agent and no subagents. His report separated three observations:

1. Task adherence: Fable stayed on track through the extended run.
2. Correctness: it still shipped bugs, including one in the launcher.
3. Cost: the session totaled approximately $600, which he described as token-efficient but still too expensive for his everyday use.

Ultra subscribers retain included Fable access after November 16, as the follow-up correction clarified. That deadline applies to the temporary inclusion offer for Pro subscribers.

Under Anthropic's launch-offer rules, eligible Ultra subscribers receive resets according to paid subscription tenure:

• Under six months: one reset.
• Six to 12 months: two resets.
• More than 12 months: three resets.

Eligibility requires paid billing before November 9, 2026 at 10:00 a.m. PT. Ultra's launch resets expire March 8, 2027 at 11:59 p.m. PT, and using one leaves the 50% Fable cap unchanged.