Avatar
Juraj 3d ago

New AI model ranking (my methodology).

KC - Kimi Code (kimi-cli)

OC - Opencode

CC - Claude Code

Tasks:

1. reverse engineering of non open source chrome extension doing advanced zero knowledge math (PeerAuth)

2. understanding and auditing solidity smart contracts and attestations (related to PeerAuth)

Kimi-cli with Kimi 2.5 surprised me, it even found something that no other model noticed.

Note - this is not coding, it wrote a paper, based on how it understood and reverse engineered code.

Reply to this note

Please Login to reply.

Discussion

Avatar
Juraj 6h ago

If you want to run Kimi-2.5 with kimi-cli through

Venice.ai (who have deployed the model really fast), you need this small change:

https://github.com/MoonshotAI/kimi-cli/pull/782

Then it works flawlessly.

So far I'm quite happy about it and it's much cheaper than the frontier models, but I don't think it's much worse.

Here's my model grading:

nostr:nevent1qgsd4dkxqewy8xum47ctpu0ltgxxsfemeewpjkdyzk9ddfcg286s0dspzamhxue69uhhyetvv9ujuurjd9kkzmpwdejhgtcqyrl55cyypkpufzq6rharze44sny597d6w7u8jml25vgartlucwreg47mtkd