Andrew Carr argues Kimi 2.5 traces make coding model distilled

Positive users hail open-weight frontier model releases as impressive feats enabling indie hackers to build agents, while negative users call distillation debates tiresome and mock training purity claims.

Based on 9 visible X reactions from 46 accounts; directional sample.

Digg Deeper

Ask a question below.

Published answers will appear here.

Commentary on X

Florian Brand@xeophon

@andrew_n_carr And basically all recent web and code data is just Claude and GPT The distillation angle is so tiresome

2:57 PM · Jul 15, 2026

RaoulDuke@RaoulDukeDegen

@teortaxesTex man this model lets indie hackers build agents without paying for apis anymore

4:00 PM · Jul 15, 2026

Michael DeMaria@demaria_michael

@gneubig @jxmnop In this situation, virtually none. It’s just an impressive feat by the team.

3:12 PM · Jul 15, 2026

Jack Morris@jxmnop

it’s still so impressive

4:53 PM · Jul 15, 2026

View all

Carl Kolon@carlkolon

@jxmnop Helpful

2:56 PM · Jul 15, 2026

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

> • Kimi distills • GLM distills • Qwen distills • Nemotron distills (Kimi & DeepSeek, which counts) you had one job…

3:22 PM · Jul 15, 2026

Graham Neubig@gneubig

@jxmnop Perhaps naive question but why do you think this is a big deal? What practical or scientific consequences does distilling or not distilling have?

2:28 PM · Jul 15, 2026

Jack Morris@jxmnop

community noted for forgetting Llama 3.1 🤦

2:54 PM · Jul 15, 2026

Graham Neubig questioned the scientific significance of avoiding model distillation.

View all

Carl Kolon@carlkolon

@jxmnop Helpful

2:56 PM · Jul 15, 2026

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

> • Kimi distills • GLM distills • Qwen distills • Nemotron distills (Kimi & DeepSeek, which counts) you had one job…

3:22 PM · Jul 15, 2026

Graham Neubig@gneubig

@jxmnop Perhaps naive question but why do you think this is a big deal? What practical or scientific consequences does distilling or not distilling have?

2:28 PM · Jul 15, 2026

Jack Morris@jxmnop

community noted for forgetting Llama 3.1 🤦

2:54 PM · Jul 15, 2026

Andrew Carr 🤸@andrew_n_carr

This is commendable and not entirely true. They use Kimi 2.5 for SFT traces, two hops is still a hop

2:45 PM · Jul 15, 2026

Jack Morris@jxmnop

ok sorry everyone apparently they did distill lol. but only a tiny bit

4:46 PM · Jul 15, 2026

Susan Zhang@suchenzang

约爱唠叨一个模型的纯血统，越傻逼混血的基因才是最厉害的