Comparing the Most POWERFUL Open Source Model versus Opus 4.8 & GPT 5.5
Channel: Chase AI
Uploaded:
Duration: PT1336S
Read a transcript preview of “Comparing the Most POWERFUL Open Source Model versus Opus 4.8 & GPT 5.5” from Chase AI, uploaded on 2026-06-22.
Transcript preview
Language: English (original, YouTube automatic captions)
[0:00] GLM 5.2 just came out this week and it
[0:02] is the strongest open-source model we
[0:04] have ever seen. And in some benchmarks,
[0:07] like you see here, even show this model
[0:09] outperforming the giants like
[0:11] Anthropic's Opus 4.8 and OpenAI's 5.5.
[0:15] But, are these benchmarks legit? How
[0:17] does this model compare head-to-head
[0:20] with Opus 4.8 and GPT 5.5? Well, that's
[0:24] exactly what we're going to answer in
[0:25] today's video as I go through multiple
[0:28] tests with these three big models and
[0:31] see how it actually performs in the real
[0:33] world. On top of that, we'll do a deep
[0:34] dive on one benchmark in particular that
[0:36] I think is rather important, as well as
[0:38] break down what we actually mean by GLM
[0:41] 5.2 being better in some instances than
[0:44] Opus and GPT 5.5. Are we talking about
[0:47] it's more efficient, it costs less, or
[0:49] it actually does better at all those
[0:50] things at the same time? So, without
[0:53] further ado, let's just hop into it.
[0:55] Now, before we jump into the
[0:56] head-to-head test, let's first look at
[0:57] some of the benchmarks that already
[0:58] exist comparing these three models. The
[1:01] one I want to really pay attention to is
[1:02] Deep Sweep. Now, Deep Sweep is a
[1:04] relatively new benchmark and it's meant
[1:06] to be an improvement upon things like
[1:08] Terminal Bench and Terminal Bench Pro.
[1:10] Now, I'm not going to go ultra deep into
[1:12] this benchmark. You can check out their
[1:13] website or their GitHub repo, which
[1:14] explains it in more detail, but it
[1:16] focuses on long-running agentic tasks,
[1:18] specifically 113 tasks across
[1:21] TypeScript, Go, Python, JavaScript, and
[1:24] Rust with isolated