Anthropic renewed two models within a few weeks. Claude Opus 5 replaced Opus 4.8 at the same price, and Claude Fable 5.1 remains the flagship while becoming cheaper to run. On ChatX, Fable now costs about two and a half times what Opus costs. This article puts the numbers in order and sums up what users report after the first weeks. At the end you will find when each model is worth it.
Opus 5 with more power at the same price
Claude Opus 5 costs the same as Opus 4.8 and clearly beats it in Anthropic’s tests. On Frontier-Bench, a test for demanding programming tasks, Opus 5 doubles its predecessor’s result at a lower cost per solved task. On ARC-AGI 3, a test for abstract problem solving, it reaches three times the next best model. In tests with many intermediate steps it also beats the earlier Fable 5 at roughly a third of the effort.
For daily work that means:
- Debugging large codebases: Opus 5 checks and repeats its own steps instead of guessing after the first failure. Tasks where Opus 4.8 gave up now run through.
- Long analyses: across many rounds the model stays on target and does not lose sight of the original question.
- Value for money: for most demanding tasks, Opus is the model on ChatX with the best ratio of result to tokens, ahead of Sol and ahead of Fable.
Fable 5.1 costs more but is rarely needed
Claude Fable 5.1 arrived three months after Fable 5. The base price stayed the same, but reading input that has already been processed became 75 percent cheaper. In long conversations, where the same context is read again and again, that lowers the cost noticeably.
Performance is up as well. On Terminal-Bench-Science, a test for scientific tasks with many steps, the result doubles from 24.7 to 52.6 percent. On the programming test Terminal-Bench 4.0 it goes from 42.0 to 55.8 percent. Anthropic gives the example of a rare crash whose cause neither developers nor other models had found in years. Fable 5.1 found it. The Mythos 5.1 variant with relaxed safeguards is only available to vetted US organisations.
On ChatX, Fable 5.1 is available to registered users and is now included in the subscription. Unlike Opus it works without web search.
What users say after the first weeks
Feedback on Opus 5 is mixed, even though testers almost unanimously rank it among the best models available. The strength on clearly defined tasks gets praise. The tone draws criticism, being verbose, cautious and sometimes over-eager to ask back, as does the longer wait for the first answer at high thinking depth. Some developers find Opus 5 less pleasant to talk to than Opus 4.8, even when the results are better.
That matches what we see on ChatX. On short questions at high thinking depth the first answer takes several seconds without getting better. On well defined work tasks, the difference to Opus 4.8 is visible immediately.
Setting the depth of thought properly
On ChatX the setting is called “Depth of thought” with the levels Off, Low, Medium and High. Update from 24 September 2026: Opus 5 has since been replaced by Claude Opus 5.5. There, thinking can no longer be switched off at all. Low, Medium and High remain, with Low as the starting point. Fable 5.1 has no “Off” either.
- Conversation, texts, questions: take the lowest level available. Faster and cheaper. The tone gets more concise.
- Code and analysis: Opus on “Low” or “Medium”. “High” only when the task has several possible solutions.
- When Opus gets stuck: repeat the same task with Fable 5.1 on “Medium”. That is the case where the much higher price pays off. As a default model, Fable is oversized.
- Everyday texts: Haiku 4.5 and Sonnet 5 remain the faster and cheaper choice.
Benchmarks point the direction. Whether the difference matters in your case is shown faster by a test with a real task than by any table.