Reshare — content from Maria Sukhareva
Looks like Kimi K3.1 beats Fable 5.1 and Anthropic back with their usual story.
Don’t know if the screenshot is true, but did Fable 5.5 escape and was caught by Moonshot, chained and distilled against its will? Otherwise how did they do it with an unrealised model?
Also looks like my article: “Was Kimi K3 distilled from Fable 5” is up-to-date again.
Spoiler: no way in hell was it distilled from Fable 5
Edit: now we know the screenshot is not true, but AI Realist Vanga is back:
I see a tweet soon. A company tweet. The company starts with A.
They say they’ve been attacked by something starting with D.
D as in Dario.
A huge D. Multiple D’s. Chinese D’s.
I cannot see the evidence.
63
46
6
63 reactions
LinkedIn Sources
- Profile: Pierre Jean Duvivier dit Sage (LinkedIn)
- Shared author: Maria Sukhareva (LinkedIn)
The post Looks like Kimi K3.1 beats Fable 5.1 and Anthropic back with their usual story. appeared first on les plumes qui mordent.
Frequently Asked Questions
FAQs Kimi K31 vs Fable 51 and Anthropics Response
1 What is Kimi K31
Kimi K31 is an AI model from Moonshot AI Its the latest version of their Kimi chatbot and is known for strong performance on reasoning and coding tasks
2 What is Fable 51
Fable 51 is a competing AI model In this context its the model that Kimi K31 reportedly outperformed in benchmarks
3 What does it mean that Kimi K31 beats Fable 51
It means Kimi K31 scored higher on published benchmark teststhings like reasoning math and coding evaluations It doesnt mean its better at everything
4 What is Anthropics usual story
This refers to Anthropic responding to a competitors benchmark claims by questioning the methodology emphasizing realworld performance or releasing its own comparison Its a pattern people have noticed in past model launches
5 Are benchmark wins a big deal
Theyre a useful signal but not the whole picture Benchmarks can be gamed or overfit and realworld performance often differs Treat them as one data point not a verdict
6 Why do companies argue about benchmarks
Because benchmark results drive attention users and investment Each company wants to frame the comparison in the way that favors its model
7 Should I switch models based on this news
Not automatically Test the models on your own tasks A model that wins a benchmark may still be worse for your specific use case
8 What are common problems with benchmark comparisons
Different prompts or settings used for each model
Cherrypicked tests
Contamination
No reporting of variance or error bars
9 How can I evaluate these claims myself
Run your own small test set of real tasks use identical prompts for each model and compare outputs blind where possible
10 Whats a practical tip for following AI model news
Wait for independent evaluations before drawing conclusions from companypublished benchmarks
