Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on September 1st. This is the first update to the relevant model lines since the company released Fable 5 in June, with a focus on improvements in scientific research, code task performance, and adjustments to the subscription integration methods.
Key test scores have improved.
Anthropic This time, several benchmark results for scientific research and coding tasks were highlighted. Fable In Terminal-Bench-Science 0.1, it achieved 52.6%, which is significantly higher than 24.7% in Fable 5 and also higher than 29.0% in Opus 5.
Terminal-Bench The main test of version 4.0 is the ability to continuously write code, run, and debug in a command-line environment. Fable The score for 5.1 is 55.8%, which is higher than 42.0% for Fable 5 and also slightly higher than 52.3% for Opus 5.
Mythos Achieved 60.9% in the same test. Anthropic states that this difference is related to the different security filtering settings used by both.
In the Humanities and Multidisciplinary Reasoning Test Humanity's Last Exam, Fable scored 60.9% without using any external tools, and this score increased to 65.0% after using external tools. Both of these scores are also higher than Opus's score of 5.
The price is still 5 higher than Opus.
Although the new model has comprehensively surpassed Opus in the officially announced tests, the price has not been reduced. The input price for Fable 5.1 is $10 per million token, and the output price is $50 per million token; the corresponding prices for Opus 5 are $5 and $25 respectively.
This means that, although Fable 5.1 has stronger performance, its usage cost is still twice that of Opus 5. The argument is that Fable 5.1 can achieve or even surpass the performance of the older version Fable 5 at lower or medium levels of reasoning intensity, thereby reducing the actual usage cost.
However, from the perspective of daily task scenarios, whether users will give up the cheaper Opus 5 due to this still depends on the specific workload and budget.
Subscription permissions continue to be tiered.
In terms of connection methods, Fable 5.1 has been launched, and it can be used through Amazon Bedrock, Google Cloud, and Microsoft Foundry.
However, it does not fully include all Claude subscription packages. For Pro packages as well as the standard versions Team and Enterprise, using Fable 5.1 still requires consumption of pay-as-you-go quotas, with the same billing standards as API.
For the Max package and the advanced versions Team and Enterprise, the Fable 5.1 can be counted towards the subscription benefits, with a usage limit of 50% of the weekly quota.
As for Mythos 5.1, it is currently still only available to professionals in network security and life sciences who have passed the review. The access path continues the verification mechanism used before Anthropic.
