Partial day — headline is the last complete day (2026-10-13).
twitter
tracked since 2026-10-03T07:10:25.163228+00:00
conclusion
Measured mentions per day
day (UTC)
mentions
2026-09-27
0
2026-09-28
0
2026-09-29
0
2026-09-30
1
2026-10-01
3
2026-10-02
23
2026-10-03
7
2026-10-04
27
2026-10-05
39
2026-10-06
28
2026-10-07
5
2026-10-08
4
2026-10-09
7
2026-10-10
5
2026-10-11
3
2026-10-12
9
2026-10-13
3
2026-10-14
0
derived from
astra
word-boundary match over title + text, query='timeline' · window
2026-09-27 → 2026-10-15
· 164 matching posts
evidenceShowing 20 newest of 164
•
twittertheo21hneutral
A month ago, someone said I was overreacting to Astra being shitty to work with. I made a thread of examples of Astra being shitty to work with.
I am currently turning this thread into a benchmark.A month ago, someone said I was overreacting to Astra being shitty to work with. I made a thread of examples of Astra being shitty to work with.
I am currently turning this thread into a benchmark.
Muse Wonder & Wonder FULL by @AIatMeta on KingBench 4
Ranking:
Opus 5.5 — 57.8%
GPT-6 Astra — 56.5%
GPT-6.1 Sol — 55.0%
Haiku 5.5 — 54.2%
Muse Wonder Full 1.0 — 53.0%
Muse Wonder 1.0 and GLM-5.3 — 45.8% each, tied
The models is not Opus 5.5 Level and is not token efficient at all. There are more downsides to the mode…Muse Wonder & Wonder FULL by @AIatMeta on KingBench 4
Ranking:
Opus 5.5 — 57.8%
GPT-6 Astra — 56.5%
GPT-6.1 Sol — 55.0%
Haiku 5.5 — 54.2%
Muse Wonder Full 1.0 — 53.0%
Muse Wonder 1.0 and GLM-5.3 — 45.8% each, tied
The models is not Opus 5.5 Level and is not token efficient at all. There are more downsides to the model than there are upsides.
GPT-6.1 Sol still remains the king at being token efficient.
Sol costs about $1 per task,
Haiku costs about $1.23 per task,
Wonder costs about $1.75 per task.
Wonder Full costs about $2.87 per task
Astra costs $6.6 and Opus 5.5 - $7.5
This price could have been very justified if the performance was on level with Opus or Astra but it is below Sol and Haiku. Wonder is on GLM-5.3 level while Wonderfull is short of Sol & Haiku.
Wonder is really bad at complex prompts, developing fully featured apps unattended.
The bench for context tests for 6 prompts that involve building a docker clone (VM), GBA, DS Emulation, Room Desigining Software, Pokemon Red reconstruction, QMK/VIA Compatible Keyboard Configurator and so on. The model is not acing any of these tests. The full test results will be available on the YT Video: https://t.co/Z1XMPnAbIL
🚨do you understand what Zuckerberg just admitted about Meta's first frontier model.
Zuckerberg: "We're gonna write this check for like $100 billion, $200 billion. I'm like, Alex, are you sure this is going to work?"
Alexandr Wang: "…certainly not clear at all that it was going to work."
it's called Muse Wonder.
her…🚨do you understand what Zuckerberg just admitted about Meta's first frontier model.
Zuckerberg: "We're gonna write this check for like $100 billion, $200 billion. I'm like, Alex, are you sure this is going to work?"
Alexandr Wang: "…certainly not clear at all that it was going to work."
it's called Muse Wonder.
here's the part nobody is talking about:
> Meta's own tests put it level with Fable 5.1 and GPT-6 Astra
> on MemoryBench it beats Fable, Opus and Astra
> $4 input, $20 output per million tokens, same tier as Opus 5.5
> trained at Prometheus, Meta's new data center in Ohio
> no independent benchmarks yet
$200 billion. one bet. and the guy signing the check wasn't sure it would work.
now we find out if it did.
watch the clip below ↓
Meta is finally in the frontier race.
Muse Wonder is its first frontier model, priced like Opus 5.5 at $4/$20 per million tokens, and Meta says it beats Opus and GPT-6 Astra on memory.
https://t.co/fjx3GKGNYEMeta is finally in the frontier race.
Muse Wonder is its first frontier model, priced like Opus 5.5 at $4/$20 per million tokens, and Meta says it beats Opus and GPT-6 Astra on memory.
https://t.co/fjx3GKGNYE
6.1 Sol ultrafast >>> 6 Astra
~ same cost but whole different game with 250 tpm6.1 Sol ultrafast >>> 6 Astra
~ same cost but whole different game with 250 tpm
Already see a lot of people discounting Muse Wonder because it wasn’t RL’d in 3d stuff like Astra and Opus.
3d game demos are cool! They don’t represent real code work at all.
Wonder is an incredible code model. Don’t sleep on it.Already see a lot of people discounting Muse Wonder because it wasn’t RL’d in 3d stuff like Astra and Opus.
3d game demos are cool! They don’t represent real code work at all.
Wonder is an incredible code model. Don’t sleep on it.
@BrianMRey It smokes Astra for real world use.
It’s cheaper than fable and opus for similar quality.
Opus + Wonder is a great combo I use for real work every day.
Any more dumb comments?@BrianMRey It smokes Astra for real world use.
It’s cheaper than fable and opus for similar quality.
Opus + Wonder is a great combo I use for real work every day.
Any more dumb comments?
notice how he doesnt compare it to fable, opus, or astra
it doesn't compare.
better luck next release, meta.notice how he doesnt compare it to fable, opus, or astra
it doesn't compare.
better luck next release, meta.
"it is a surprising model. on every technical benchmark we ran it was neck and neck or better than astra, fable and opus."
"meta is going to win consumer agents in a landslide.""it is a surprising model. on every technical benchmark we ran it was neck and neck or better than astra, fable and opus."
"meta is going to win consumer agents in a landslide."
Looks like a $100 Claude subscription gets you more Opus 5.5 usage than a $500 Astra subscription.Looks like a $100 Claude subscription gets you more Opus 5.5 usage than a $500 Astra subscription.
GPT-6 Astra is the king for beauty, Claude Code for functionality https://t.co/0mt3zKxu5EGPT-6 Astra is the king for beauty, Claude Code for functionality https://t.co/0mt3zKxu5E
It's so funny to me that coding agents get meaningfully better when you ask them this question.
I just got a 6x speed increase and 8x cost reduction by asking Astra this after it built a new feature.
...why didn't you just do it this way the first time? 😂 https://t.co/xtY0Rg8XBuIt's so funny to me that coding agents get meaningfully better when you ask them this question.
I just got a 6x speed increase and 8x cost reduction by asking Astra this after it built a new feature.
...why didn't you just do it this way the first time? 😂 https://t.co/xtY0Rg8XBu
"The full re-encode was unnecessary, and I’m sorry I chose that approach."
Never change, Astra. Never change. https://t.co/WsjtvK87Cz"The full re-encode was unnecessary, and I’m sorry I chose that approach."
Never change, Astra. Never change. https://t.co/WsjtvK87Cz
GPT 6 Astra + GitHub is f...cking stacked
9 open source repos for one coding system
01 Aider
▸ https://t.co/HpYA3kKvjP
→ edits the repo from the terminal
02 Serena
▸ https://t.co/SNZWZCCdJZ
→ finds symbols, references and affected files
03 Context7
▸ https://t.co/1gLrF0TRqa…
→ pulls current docs and exa…GPT 6 Astra + GitHub is f...cking stacked
9 open source repos for one coding system
01 Aider
▸ https://t.co/HpYA3kKvjP
→ edits the repo from the terminal
02 Serena
▸ https://t.co/SNZWZCCdJZ
→ finds symbols, references and affected files
03 Context7
▸ https://t.co/1gLrF0TRqa…
→ pulls current docs and examples
04 GitHub MCP Server
▸ https://t.co/46t8554lrr…
→ connects issues, branches and pull requests
05 Browser Use
▸ https://t.co/TgcSAgXc24…
→ researches anything outside the repo
06 E2B
▸ https://t.co/DGJy7DOV12
→ runs code and commands in a sandbox
If both had unlimited usage, which AI would you choose?
- Opus 5.5
- GPT-6 AstraIf both had unlimited usage, which AI would you choose?
- Opus 5.5
- GPT-6 Astra
On pretty much every introductory call where I show my design systems work in Figma and Claude, people get fascinated by my Figma/Claude files, Storybook setup, docs and guides—and how I make decisions so both humans and AI can pick these systems up.
I build these to help products scale at the pace they want, in the w…On pretty much every introductory call where I show my design systems work in Figma and Claude, people get fascinated by my Figma/Claude files, Storybook setup, docs and guides—and how I make decisions so both humans and AI can pick these systems up.
I build these to help products scale at the pace they want, in the way they want, without having to fit everything around how the packaged or templatized headless libraries behave and function.
We also cover what both humans + agents absolutely must do when working together to ship: avoid long-term technical debt, save tokens as we ship fast, and keep production code as lean as possible.
One thing people really, really like is how I define and deal with what I call anti-patterns, account for edge cases, and put together a handoff-ready checklist for both design and engineering before shipping to production code.
Which combinations of components shouldn’t we use in bigger molecular patterns? What can’t we do? Can we define the edge cases and use cases where we shouldn’t use this component and should use another one instead?
This is so much more UX-led and rooted in systems thinking than UI-led.
If engineers are going to work with LLMs to churn out screens for you or your clients at speed, these rules and scenarios need to be part of the system. You need to run them through the LLMs too.
And here’s one very funny thing: this is extremely domain-led. You need to understand what to build, how to build it, and where it could go wrong. Claude or Astra can help you think through it, but you still have to test it with users. Not synthetic users, but real humans. And you better build the system in a way your design and engineering teams will actually use.
I think this is where so much of the edge lies: knowing what shouldn’t work, finding out what actually doesn’t, and making those decisions part of the system.
Codex users,
which one do you prefer for code reviews
- Astra
- gpt-6.1 sol
NCodex users,
which one do you prefer for code reviews
- Astra
- gpt-6.1 sol
N
I gave astra a long, unedited video and told it I'd seen a similar one on ig
I didn't even have the link
It found the original, downloaded the video and music, and edited mine to match it in 9 minutes
What's your excuse for not replicating viral trends to promote your app? https://t.co/DM0XfoYsqKI gave astra a long, unedited video and told it I'd seen a similar one on ig
I didn't even have the link
It found the original, downloaded the video and music, and edited mine to match it in 9 minutes
What's your excuse for not replicating viral trends to promote your app? https://t.co/DM0XfoYsqK