Kamil
66 posts
Kamil
@madebykamil
Founder @SmophyAI - building an AI workspace that turns multiple models into real work, not just chat.
Katılım Haziran 2026
28 Takip Edilen23 Takipçiler
Gemini 3.5 Pro just missed its third deadline. Google scrapped the whole base model and rebuilt from scratch, still not enough.
GPT-5.6, Grok 4.5, and DeepSeek V4 all shipped this month.
Prediction markets now say August 7. Being early isn't the hard part. Shipping is.
English
come for the best model, stay because we don’t treat you with contempt
English
@finkd Ran the numbers: on SWE-Bench Pro, Muse Spark 1.1 gives 14.47 score per dollar of output tokens. Grok 4.5 gives 10.78. Opus 4.8 gives 2.77.
Full breakdown here: x.com/madebykamil/st…
Kamil@madebykamil
Grok 4.5 was the cheapest near-frontier model from a major US lab. It held that for one day. Meta took it. SWE-Bench Pro, score per dollar of output tokens: Muse Spark 1.1 - 14.47 Grok 4.5 - 10.78 Opus 4.8 - 2.77 GPT-5.5 - 1.95 xAI and Meta published these tables separately. Both put Opus 4.8 at 69.2 and GPT-5.5 at 58.6, which is what makes the comparison worth running. Meta leads tool use. Opus leads SWE-Bench Pro. GPT-5.5 leads the other coding benchmarks. Nobody wins everything.
English
(1) Today we're releasing Muse Spark 1.1 -- a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.
English
Grok 4.5 was the cheapest near-frontier model from a major US lab. It held that for one day. Meta took it.
SWE-Bench Pro, score per dollar of output tokens:
Muse Spark 1.1 - 14.47
Grok 4.5 - 10.78
Opus 4.8 - 2.77
GPT-5.5 - 1.95
xAI and Meta published these tables separately. Both put Opus 4.8 at 69.2 and GPT-5.5 at 58.6, which is what makes the comparison worth running.
Meta leads tool use. Opus leads SWE-Bench Pro. GPT-5.5 leads the other coding benchmarks. Nobody wins everything.
English
𝕏 is a great platform for product announcements, especially if done by the CEO directly. Way more interesting to the public than generic press releases.
This post by Mark Zuckerberg already received over 12 million views for free!
Mark Zuckerberg@finkd
(1) Today we're releasing Muse Spark 1.1 -- a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.
English
Grok 4.5 is 1st on intelligence per dollar. By 4x.
9.00 vs Opus 4.8 at 2.24, GPT-5.5 at 1.83, Fable 5 at 1.20.
On SWE Bench Pro it scores 64.7% to Opus 4.8's 69.2%, but costs $0.096 per task vs $1.68. 17.5x cheaper.
Frontier intelligence stopped being a spending contest.
English
@madebykamil You may not like Musk, but you have to admit that he is incredibly hardworking and stubborn in pursuing his goals. Madman 👏
English
@RohanJ_Markets @SwanDesk We’ll see what it looks like in a year or two. Healthy competition is good, it pushes everyone to constant development. We’ll see where the border is.
English
@madebykamil @SwanDesk Once the ground work is there, the process becomes faster.
Incredibly excited to see what they all have coming. Every day there’s something insane
English