Senior software engineer testing open-source, self-hosted alternatives to paid AI agents and dev tools. Hands-on notes on cost, limits and what holds up.
Goiás, Brasil
Joined April 2020
- Tweets273
- Following97
- Followers112
- Likes512
Pinned Tweet
Grok Bot alternatives #2: @agentes_chat.
A strong contender with features on another level compared with the competition. My early tests have worked really well. They've also announced an open-source, self-hosted version.
Stênio Aníbal retweeted
Nice DeepSWE score! Does it survive a lowercase filename?
Introducing MaxxBench: 89 DeepSWE tasks with small changes to the instructions or starting code. Same engineering problem, slightly different requirement.
Models clustered around 74% on DeepSWE end up 23 pp apart.
Good at coding, or good at DeepSWE? 🧵
Stênio Aníbal retweeted
AA shows Sol is faster than Opus on high
Frontier Code says that Opus is faster, and so does real usage
A real wall clock time metric from AA would be great.
Stênio Aníbal retweeted
Everyone is discussing what a harness is. Nobody shows you how to build one. And everyone can.
So, I built my own Claude Code: github.com/phuryn/mini-harne…
It's called Mini Harness: about 1,200 lines of Python.
Other minimal harnesses exist, but none has /goal mode, permissions + auto mode, skills and MCP servers. I wanted a fully working one to explain how a harness is built, down to Anthropic's server-side compaction.
Surprisingly, it became the top harness on Claude Haiku.
How it works: 🧵
“Bring your own API key” sounds wonderfully cheap until you discover you're paying for the app, the model, and the automation connecting them.
I'd rather see the cost of one real workflow than another pricing page advertising $0/month.
Stênio Aníbal retweeted
On 11/11 we're opening the chat where your team and AI agents work together.
Your team asks. Agents investigate, ask each other for help, and get it done. You approve what matters.
agentes.chat
Stênio Aníbal retweeted
Anthropic finished three models since July. You've used none of them.
METR's report out today calls them Model 2, 3 and 4. Those are codenames. My bet on the real names:
Model 2 (done July): Fable/Mythos 5.5
Model 3 (done Aug to Sep): Opus 6
Model 4 (done Sep to Oct): Fable/Mythos 6
METR says Anthropic "is holding onto more recent models for slightly longer before their public release."
Fable 5.5 is coming. Fable 6 is the one they won't give us.
You should check out @agentes_chat. These guys are making some HUGE promises.
I've played around with it a bit, and the potential looks massive.
Stênio Aníbal retweeted
I use Opus Medium as a default smart model
I use Sol High when I'm doing computer configuration stuff across devices or computer use
I use Sol Low when I'm doing a very quick and simple computer task that's unlikely to get messed up
I use Sonnet Medium when I'm running low on my Claude limit but need some Claude vibes for a visual output
I use Opus Extra High for my hardest problems
Stênio Aníbal retweeted
Here's the cost speed capability triangle as an actual 3D chart
By the way, fun Easter egg: if you hit the D key 3 times on Priors, it takes you to a 3D view
Stênio Aníbal retweeted
Mistral may have trash models, but Vibe subsidizes GLM 5.3 way better than Z.ai does
