$200 Claude Ultra: Opus Can Generate $15,000, Fable Only Worth $3,000

Theo (t3.gg) posted a set of usage data for the Claude Ultra plan on X. For a $200/month subscription, if used entirely with Opus 5.5, converted at API prices, it can produce about $15,000 worth of usage per month; switching to Fable 5.2, only about $3,000 remains. He initially wrote Fable as 5.5, then corrected it to 5.2.

Someone replied "Spot on" and posted an image in agreement.

Fable also has another limit: at most only 50% of the weekly quota can be used. If you use only Fable for a week, the other 50% of the quota goes unused.

Theo's analysis has documentation. He said he logged all requests in Claude Code and analyzed more than 1.1 million real requests sent to the Anthropic API. The comparison dimensions include dollars and percentages, tokens and percentages, the impact of the 50% quota limit, and how much work can be completed per 1% of quota.

Calculated by "how much work gets done per week," the picture is more complicated. Using only Opus, about 1,200 conversation turns and 230 threads can be run per week; using up Fable quota first and then switching to Opus gives about 900 turns, with slightly more threads. Theo himself put a question mark on "slightly more." His impression is that Fable is more likely to finish the job within a single prompt, but he also cautions that a large amount of recent usage was spent on running benchmarks, so the numbers will still change.

On pricing, a reply pointed out that Fable 5.2 is cheaper than 5.1: input price dropped from $10 to $8, output price from $50 to $40, and model efficiency also improved, using fewer input tokens for the same result. Theo posted a comparison chart. Being cheaper than the previous generation and being cheaper than Opus are two different things.

The disagreement in the discussion is also here. Some netizens suggested using Fable as a dispatcher or review model while still using Opus as the main model; some felt Opus 5.5 is already good enough, leaving Fable only for review and hard-to-debug bugs. Others joked that instead of agonizing over quotas, just open two Claude subscriptions.

The version number mix-up was corrected by many people. When Theo tweeted, he wrote Fable 5.5; replies kept reminding him it was Fable 5.2. He then corrected himself and complained that this would be annoying. Someone in the comments also posted a video.

These are still early numbers. Theo said he will continue to update. For ordinary users, subscription quotas should not be judged only by how strong the model is; unit price, weekly limits, and completed work should all be calculated together. For developers, logging your own requests and then doing the math is more reliable than looking at the marketing page.

Source: Theo (t3.gg)'s X post: https://x.com/theo/status/2119960608816988596