September 1, 2026
Fable 5.1 is here. The 20x plan is still not 20x.
Anthropic shipped Claude Fable 5.1 today at the same token price and a much cheaper cache. The subscription math underneath did not move, and the 50 percent boost ends in two weeks.
Anthropic released Claude Fable 5.1 this morning, together with Mythos 5.1 for the vetted list. Fable is the one you can use. It is in Claude.ai, in Claude Code and on the API as claude-fable-5-1.
In June Fable 5 was pulled within days. In July it came back and became my default next to Sol. A point release that ships to everyone on day one is already progress.
What changed
The numbers Anthropic leads with: 55.8 percent on its agentic coding benchmark, up from 42.0 for Fable 5. Mythos 5.1 sits at 60.9 on the same test. Terminal-Bench-Science doubles, 24.7 to 52.6. Vendor numbers on a vendor benchmark. I will know in a week whether it holds up in a repo.
The API price did not move: $10 in, $50 out per million tokens. Cache reads dropped to $0.25 per million, 75 percent less than before. Anthropic says that works out to around 25 percent cheaper for typical workloads and up to 45 percent for long agent runs, because those runs are mostly cache reads. That part I believe. An agent that rereads the same repo all afternoon is a cache bill.
The subscription math
This is the part that did not change, and it is the part most people pay for.
Fable 5.1 is included on Max and Team Premium, capped at 50 percent of your weekly limit. Pro gets it through usage credits only. The one-time $100 credit that came with Fable 5 does not repeat for 5.1.
Then there is Max 20x. It costs $200 and Max 5x costs $100. The name promises four times the usage. The pricing page says “20x more usage than Pro”.
Anthropic’s own weekly figures, published when the weekly limits arrived last summer:
| Limit | Max 5x | Max 20x | Ratio |
|---|---|---|---|
| Five-hour session | 5x Pro | 20x Pro | 4x |
| Sonnet per week | 140 to 280 h | 240 to 480 h | about 1.7x |
| Opus per week | 15 to 35 h | 24 to 40 h | about 1.3x |
The 4x holds inside a five-hour window. Across a week, which is the limit that runs out, the $200 plan buys about 1.7x the Sonnet and about 1.3x the Opus. A class action filed in California in June is about exactly this gap.
And the weekly number is about to shrink. The 50 percent Claude Code boost that has been running since May ends on September 13. Anthropic’s replacement is a permanent 25 percent over the old base. In Anthropic’s own words, that is a 17 percent reduction compared with today.
My read
Fable 5.1 is a better model at the same price with a cheaper cache. On the API that is a clean win.
On a subscription the story is the same as it has been all summer. The multiplier on the box is a session number. The ceiling is a weekly one, and it goes down in two weeks. If you took the 20x plan because you wanted four times the week, you are paying an extra $100 a month for somewhere between 1.3x and 1.7x. Run /usage before you renew.
