I tested Ox Alpha vs Kimi K3 Ran Ox Alpha through the Claude Code harness. Honestly, I expected much more from Ox Alpha. IMO, it wasn't even close to GPT-5.6 Terra. Kimi K3 was the clear winner for me. What model could this be?

Kimi K3 much better than alpha

Kimi is so goated man

It is indeed

yeah its not fronteir level it is more like opus 4.8 level

What’s that rocket prompt

Kimi K3 is goat 🔥 What prompt did you use ?

Kimi is really good

is building games is benchmarks for llms now?

@Bhavani_00007 interesting, didn't think Kimi K3 would shine like that. Makes me wonder what Ox Alpha's main focus actually is. Maybe different strengths?

I love kimi! Can’t be beat

So Ox Alpha was just benchmaxxed?

Kimi was always a strange beast that in some cases performed exceptionally well.

Running both through the same harness is what makes it hold, otherwise its two variables

the honest answer is that kimi k3's performance might just be a result of the specific tuning done on the claude code harness, not necessarily an inherent strength of the model itself. could be worth exploring more context around the experiment setup.

Kimi K3’s results are genuinely impressive, but a benchmark should show more than just the final demo. To judge the result fairly, we also need the full context: how long it took, how much compute or how many tokens were used, and how many external assets, images, 3D models, or other materials were involved.

grok 4.5

Harness fit is the hidden variable here - Claude Code's scaffold is tuned for Claude-style behavior, so Ox Alpha may be fighting the harness, not the task. Vibes can't separate model from scaffold; behavioral evals can. #LLM #LLMOps @Kimi_Moonshot

Kimi for me

I made a comparison of Grok 4.6 vs CHATGPT 5.6 Sol

gemini pro? any chance sharing the prompts? :) Great work

How do you tested?

Ox Alpha's performance gap suggests the market hasn't priced in its true potential yet.

So ox alpha is not haiku then composer 3 or Qwen 3.8 flash next

nteresting result. My take: the hype around Ox Alpha is ahead of the evidence. It may be useful for long-running coding tasks, but reports on speed, reliability, and output quality are mixed. Still worth trying while it’s free until August 27.

Same-harness comparisons are the only fair kind, nice. Since winners rotate monthly, we keep the harness open source and model-agnostic — swap what's underneath without rebuilding agents or memory: #AI #AIAgents #Agentlas

Kimi wins by fatality

There is a huge gap between them.

Kimi is fable level frontend

What I would separate is model quality from harness reliability. Did Ox Alpha lose because the output was worse, or because the run stopped early / needed more steering? The same model can look very different if the execution loop is flaky.

one harness run can expose a lot, but it can also bake in workflow bias. the real signal is whether Kimi K3 keeps winning across tasks, budgets, and agent loops

the hidden issue is that code harnesses reward different failure shapes. Kimi often looks better when the task needs stubborn iteration, while mystery models can look impressive until the repo starts pushing back.

K3 if you also add 2 more for compare Opus and sol its easy to see who gonna win

听着像是高期待开局，结果被 Kimi K3 截胡了。

名字带Alpha的容易翻车，K3这种低调选手反而能打。

kimi failed the rocket test. the rocket flew off with the harness.