Fable 5 or Fable 5.1: upgrade or wait
If you are building an agent that resends the same context, upgrade: the price is identical and the cache read is now a quarter. If you write short pieces of code, waiting costs you nothing; the gap on that work is under three points.
Specifications where both sides have a verified number
| Claude Fable 5 | Claude Fable 5.1 | |
|---|---|---|
| API model ID | claude-fable-5 | claude-fable-5-1 |
| Context window | 1 million tokens | 1 million tokens |
| Max output | 128k tokens | 128k tokens |
| Input price | $10 per million tokens | $10 per million tokens |
| Output price | $50 per million tokens | $50 per million tokens |
| Cached input price | $1 per million tokens | $0.25 per million tokens |
| Knowledge cutoff | January 2026 | June 2026 |
| Where it ships | Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, Microsoft Foundry | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS |
A row where one side has no verified number is left out entirely, because readers read a half-empty row in favour of the filled side.
Only one number in the table above is different
Input is ten dollars on both, output is fifty on both, the context window is one million tokens on both and the output cap is 128K on both. The only column that moved is the cache read: one dollar on Fable 5, twenty-five cents on Fable 5.1.
Anthropic calls that an exception itself. Cache reads cost 0.1x the base input price on every Claude model, and 0.025x on Fable 5.1 and Mythos 5.1 alone. In the whole price list, that is the single row that steps outside the rule.
So who is the upgrade worth it for
Anyone resending the same text over and over. An agent resends the same system prompt and the same files on every turn, and with caching on, most of its bill is the column that just fell to a quarter. Anthropic states roughly 25 percent lower cost on typical workloads and up to roughly 45 percent on agentic ones.
And who it is not worth it for: anyone sending one long call with no repetition. There is nothing to read from cache there, and you pay the Fable 5 price exactly. In that case the upgrade is not cheaper. It is only better.
The capability jump is not uniform either
Anthropic published three evaluations with a figure for both models. Terminal-Bench-Science goes from 24.7 to 52.6, Terminal-Bench 4.0 from 42.0 to 55.8, and CursorBench from 70.5 to 73.4.
The longer and more multistep the task, the wider the gap. On CursorBench, the closest of the three to everyday coding work, the difference is two point nine points. If your work is writing short pieces of code, this upgrade gives you almost nothing.
One thing the table does not show and you should know
Three changes in this version are breaking: forced tool use returns an error, earlier models cannot read this model thinking blocks, and editing an earlier turn invalidates the thinking blocks. Swapping the model ID and deploying without reading those three costs you the first day.
And one caveat over every figure above: all six benchmarks Anthropic published are self-reported, with no independent leaderboard behind them. The only third-party measurement today is the Artificial Analysis index.
Pick this if Claude Fable 5
- Your workload has no repetition, so nothing is ever read from cache and the upgrade takes nothing off your bill.
- You run an architecture that swaps models mid-conversation; earlier models cannot read 5.1 thinking blocks.
- You depend on forced tool use, which returns an error on 5.1.
Pick this if Claude Fable 5.1
- You are building an agent that runs for hours and resends the same system prompt and files every turn.
- Your work is research or multistep scientific work, which is exactly where the largest jump lands.
- A fresher cutoff matters to you: 5.1 reads through June 2026 against January 2026.
Questions people actually ask
What is the difference between Fable 5 and Fable 5.1
Input price, output price, context window and output cap are all four identical. The differences are these: cache reads went from one dollar to twenty-five cents, the knowledge cutoff moved from January to June 2026, and benchmark scores rose most on long agentic work.
Does upgrading to Fable 5.1 cost less
Only if you read from cache. The base price did not change at all, so every saving comes from the cache column. Anthropic states roughly 25 percent on typical workloads and up to roughly 45 percent on agentic ones. On one long call with no repetition the saving is zero.
Is swapping the model ID enough
No. Three changes are breaking: forced tool use returns an error, earlier models cannot read 5.1 thinking blocks, and editing an earlier turn invalidates those blocks. Read those before you deploy.