Kimi K3 matches Claude Fable 5 on DeepSWE at 35% cost

Together Compute found Kimi K3 leads at higher pass@k.

Combined views

3.3K

2 posts, first seen 11h ago

10 likes

Kimi K3 matches Claude Fable 5 on DeepSWE at 35% cost

Together Compute found Kimi K3 leads at higher pass@k.

Reactions from ranked influencers

2 posts

CP Chris Paxton@chris_j_paxton

1:10 PM · Jul 18, 2026

854TECH

Higher pass@ should correspond to an actual software workflow with a human somewhat in the loop right. So another big data point

Together AI@togethercompute

We analyzed Kimi K3 vs. Claude Fable 5 for software engineering tasks using DeepSWE. Kimi K3 gets you the same performance as Fable 5 at ~35% of the price, and it actually pulls ahead at higher pass@k's. More insights in the thread! https://twitter.com/ZainHasan6/status/2078289171807150312

TA Together AI@togethercompute

6:26 PM · Jul 17, 2026

Zain@ZainHasan6

Deepdive: Kimi K3 vs. Fable 5 on software engineering/DeepSWE tasks. The open frontier isn't 6 months behind anymore! full deep-dive 👇(1/n)🧵

Combined views

3.3K

2 posts, first seen 11h ago

10 likes comments

2 comments

172 reposts

Sentiment

Positive82.8%17.2%Negative

Positive Read

Many users are excited about Kimi K3 matching Claude Fable 5 on software tasks at 35% cost because it delivers strong performance with major savings, while some dismiss the results as limited benchmarks or question their reliability.

Based on 36 visible X reactions from 80 accounts.

Digg Deeper

Ask a question below.

Published answers will appear here.

Reactions from the X community

WA Walter@Walterqwix

7:19 PM · Jul 18, 2026

@togethercompute sorry i dont believe any of it, i tested GLM 5.2 a month ago and it was complete crap, useless garbage, for something like Fable to arise there must be in between stages, i haven't heard anything even like opus 4.6 from chinese, this has to be some sort of scam/hoax

OK Omar K@bridgetoai

1:23 PM · Jul 18, 2026

@chris_j_paxton Exactly right. Higher pass@k is where my fintech workflows live. I still need that human review layer for regulated code, so 35% cost changes everything.

R(Ramesh (CV) Chitor@rameshchitor

10:46 PM · Jul 18, 2026

@togethercompute These are some really good results. Thanks for sharing @togethercompute

F(Florian🪶 (httpz)@zktrivo

12:39 AM · Jul 18, 2026

@togethercompute what a time to be alive!

View all

VV Viral Vora@viralvora_

8:45 PM · Jul 18, 2026

@togethercompute 🤯🤯

CP Chris Paxton@chris_j_paxton

1:10 PM · Jul 18, 2026