Snapper AI

Building AI agents, tools & tutorials on Claude Code, Cursor, and Codex. Free templates + newsletter:

[Snapper AI @SnapperAI]

GLM-5.1 is the best model to use on Hermes, and ranks very highly on OpenClaw too.

I built a new benchmark to test models on OpenClaw and Hermes, testing persistent memory, tool discipline, protocol compliance, and injection safety.

I put 11 models through the benchmark, including GLM-5.1, GPT-5.4, Kimi K2.5, Grok 4.20, and more.

Check out the results below!

https://t.co/98CQBuKBoz

[Snapper AI @SnapperAI]

Grok 4.3 just became the strongest all round model in my benchmark set.

Ran it alongside GPT-5.5, DeepSeek V4 Pro and Qwen 3.6 Max across coding, OpenClaw and Hermes.

It was the only model from this update that held up across all three benchmark families.

Full rankings →

[Snapper AI @SnapperAI]

First look at how Kimi K2.6 did on my Coding Benchmark and OpenClaw and Hermes runtime fit benchmark. Solid results!

https://t.co/rFcZqdxSm5

[Snapper AI @SnapperAI]

I tested Opus 4.7 across three custom-built benchmark: Coding, Multi-Turn tasks, and OpenClaw & Hermes runtime fit.

Looks like a very strong model for OpenClaw, full results here: Opus 4.7 Ranked on OpenClaw, Hermes & Coding Benchmarks vs 12 Models
https://t.co/ZSTqaCFjDV

[Snapper AI @SnapperAI]

@ASvanevik GLM-5 or 5.1? On my benchmark GLM-5.1 came out on top, beat out Opus and every other model: https://t.co/tFzMSGE5L6

[Snapper AI @SnapperAI]

@bcherny Loving the update, will be using this a lot more, starting to prefer it over CLI

[Snapper AI @SnapperAI]

@AlexFinn Agree using both is the way to go. They both have different strengths and nuances. Those differences showed up in my benchmark where I tested to see which models work best on each: https://t.co/je9S8KIUOq

[Snapper AI @SnapperAI]

@BLUECOW009 Very good. It ranked #1 overall on my Hermes/OpenClaw benchmark: https://t.co/xBAv3ugMIf

[Snapper AI @SnapperAI]

@NousResearch @anthonyronning GLM-5.1 came out on top in my Hermes benchmark, I’ll be doing a full setup and migration to GLM-5.1 next video: https://t.co/PFsBHHj9GS

[Snapper AI @SnapperAI]

@anthonyronning @NousResearch Yep GLM-5.1 came out on top in my benchmark: https://t.co/PFsBHHj9GS

[Snapper AI @SnapperAI]

@moltbanker @AlexFinn I’ve only built the baseline benchmark so far, so no long running memory tasks yet. But I’ll build that into future advanced versions.

[Snapper AI @SnapperAI]

@0xShayan @Teknium I built a benchmark to test all models on Hermes runtime and GLM-5.1 came out on top.. it’s a solid model: https://t.co/3Q7cPXwgFZ

[Snapper AI @SnapperAI]

@AIHacksByMK @AlexFinn For the benchmark I run all models via API. It’s the cleanest way to measure token usage, cost, wall time while keeping env same for all models.

[Snapper AI @SnapperAI]

@rileybrown GLM-5.1.

I built a Hermes/OpenClaw benchmark and tested 11 models: https://t.co/H9wUfW2i2q

[Snapper AI @SnapperAI]

@bcherny Recorded a video walkthrough on how to set up most of the features! https://t.co/dfVHC7oxof

[Snapper AI @SnapperAI]

Recorded a video walkthrough of how to set up most of @bcherny's favourite features from this thread!

https://t.co/dfVHC7oxof

[Boris Cherny @bcherny]

I wanted to share a bunch of my favorite hidden and under-utilized features in Claude Code. I'll focus on the ones I use the most.

Here goes.

[Snapper AI @SnapperAI]

https://t.co/40ACO7PiAA

[Snapper AI @SnapperAI]

@startupideaspod Video here on how to set it up: https://t.co/dH09mY0OVL

[Snapper AI @SnapperAI]

@trq212 I made a quick video on it, very easy process! https://t.co/N2R3PtTkNn