Strag (@Strag_AI) header
Strag (@Strag_AI) profile picture

Professional AI yapper. Thoughts & opinions are my own.

Philippines
Joined January 2022
Photo by Strag (@Strag_AI)Photo by Strag (@Strag_AI)Photo by Strag (@Strag_AI)Photo by Strag (@Strag_AI)Photo by Strag (@Strag_AI)Photo by Strag (@Strag_AI)Photo by Strag (@Strag_AI)Photo by Strag (@Strag_AI)
Replying to @ComfyUI
putting a faster seedance model in the same graph people already run is how it actually gets used, not as a one-off demo
8
this is how a new model actually gets used. drop it into the same router as the rest of the stack and compare it on real jobs
1
Replying to @MatthewBerman
that's the distinction that matters. the api is what we can run today, and the smaller open weights drop is the one to watch next
3
Replying to @maria_rcks
agreed. for daily work it's still the one i keep coming back to, even when a newer model looks stronger on paper
5
Replying to @Xianbao_QIAN
a cheap flash model that will be open sourced is more interesting than another closed mid-tier. the free window is the time to see if the quality holds
1
5
Replying to @opencode
a two-week unlimited window is the right way to test a fast model. run it on real work before deciding if it stays in the stack
25
Replying to @_avichawla
this is the right direction. once a task keeps repeating, the model should only step in where judgment is actually needed
11
Replying to @testingcatalog
leaderboards move fast. i'll wait to see how it holds up on messy, long agent jobs before i move anything over
2
Replying to @jarrodwatts
the limits always run out before the work does. a third sub is starting to feel like the normal setup
33
the coordinator still makes me check the threads myself. i want it to flag when something stalls or is waiting on me, without me having to hunt for it.
We've made some big updates to Projects in Claude Code! In a project, Claude coordinates your tasks across parallel threads. The coordinator is now much more proactive. It answers your questions in the project chat and gives you updates on the status of your ongoing threads.
21
Replying to @moritzkremb
got it. being able to rerun the same pass later is what makes it worth keeping
5
Replying to @_TK_YM_
yeah. this is the kind of change you only notice once a job runs for a while
1
8
Replying to @NabhaniMehdi
that's fair. i'd try it on reads first and keep writes serial until it's stable
7
Replying to @kimmonismus
source?
19
a whole swimmable world in one tiny html file. no models, no assets. that's a lot of ocean for 97kb 👀
Fable 5.2 built a swimmable underwater world in a single 97 KB HTML file. One of the most impressive examples I've seen so far.
19
claude code shipped an experimental mode (off by default) where it writes a little javascript to run several tool calls at once, instead of bouncing back after every step. the chatty loop was the tax. this is the right direction, even if it's still early.
2
1
3
152
if this really is the next flash model, the cheap tier just got more interesting still unconfirmed, but google is moving through the periodic table pretty fast
Gemini 4 Oxygen (the next flash model) Left is xhigh Hydrogen, the right is what was routing in Arena, same prompt and there is a decent difference
13
the jump on real paid freelance work is hard to ignore, at least for jobs that start with a clean brief. what's left is the messy part: turning a vague call into that brief, then checking the output is actually usable.
Talking about acceleration: In 13 months, the best AI agent went from completing 2.5% of real freelance projects on the Remote Labor Index to 41.3% with Claude Fable 5.2. These were paid projects that took human professionals 29 hours on average, with briefs provided and client calls excluded. So yeah, white collar work will get a wake-up in 2027.
9
if you're on claude pro, included fable 5.2 access runs through tomorrow night, nov 16 at 11:59pm pt, up to half your weekly limit. after that it goes back to usage credits. if you've been saving a long agent job for it, that's the window.
24