For the past year, every serious AI conversation has circled back to the same dependency: Nvidia.
If you wanted frontier performance, you needed their chips, If you wanted scale, you needed more of them.
Then GLM-5 dropped & suddenly, benchmark charts that usually move inch by inch started shifting.
There’s also a growing buzz online claiming GLM-5 may have been trained independently of Nvidia hardware, some even speculate about alternative stacks like Huawei’s. Nothing official confirms that. But the fact that people are even asking that question tells you how disruptive this release feels.
Because the real reason people are talking isn’t just the size. It’s what GLM-5 is capable of.
It is designed for longer, more demanding tasks where the model has to think in steps, plan ahead, and stay consistent instead of just giving a clever one-shot answer.
It can handle multi-step workflows. It doesn’t lose track halfway through long contexts. And on Vending Bench 2, it ran a simulated business for an entire year and ended with a $4,432 balance.
I’ve seen plenty of open models get close to the big closed systems before. But rarely do they feel balanced across everything.
GLM-5 is one of the first open models in a while that doesn’t feel “almost there.”
It feels like it’s actually in the same arena.
And that’s why it’s suddenly everywhere.
GLM-5 by the Numbers: Why the Charts Are Turning Heads
If you’ve looked at the Artificial Analysis or BrowseComp charts this week, you’ve seen GLM-5 at the top of the open-weight list. Here is why:
744 Billion Parameters
This is a massive Mixture-of-Experts (MoE) model.
Only 40B parameters activate per token, which keeps it efficient while still operating at frontier scale.
It’s big but it’s also designed to be practical.
28.5 Trillion Training Tokens
That’s a serious training run.
For context, token count matters because it directly affects how much structured knowledge and pattern exposure a model absorbs.
77.8% on SWE-bench Verified
That puts it firmly in the serious coding category.
SWE-bench tests real-world software engineering tasks. Scoring in this range means GLM-5 isn’t just generating pretty code. It’s solving structured problems.
$1.00 per 1M Input Tokens
Pricing is where things get disruptive.
At roughly five times cheaper than top-tier closed models, GLM-5 suddenly becomes interesting for startups and builders. Cost changes adoption.
MIT License + Open Weights
You can download it.
You can deploy it.
You Handle The Infrastructure Cost As well.
When you stack all of that together, It looks like a serious contender.
And that’s why this week, the charts aren’t just updating.
They’re shifting.
GLM 5 Vs AI Giants
Let’s put hype aside and look at the scoreboard.
According to benchmark data published by Z.ai, GLM-5 is competing directly with models like:
- Claude Opus 4.5
- Gemini 3.0 Pro
- GPT-5.2
- DeepSeek-V3.2
- Kimi K2.5
Reasoning Benchmarks
| Benchmark | GLM-5 | Claude Opus 4.5 | Gemini 3.0 Pro | GPT-5.2 | DeepSeek-V3.2 | Kimi K2.5 |
|---|---|---|---|---|---|---|
| Humanity’s Last Exam | 30.5 | 28.4 | 37.2 | 35.4 | 25.1 | 31.5 |
| Humanity’s Last Exam (w/ Tools) | 50.4 | 43.4* | 45.8* | 45.5* | 40.8 | 51.8 |
| AIME 2026 I | 92.7 | 93.3 | 90.6 | — | 92.7 | 92.5 |
| HMMT Nov 2025 | 96.9 | 91.7 | 93.0 | 97.1 | 90.2 | 91.1 |
| IMOAnswerBench | 82.5 | 78.5 | 83.3 | 86.3 | 78.3 | 81.8 |
What this tells us:
GLM-5 isn’t dominating every reasoning test, but it consistently lands in the same tier as frontier closed models & sometimes outperforms them.
Coding Performance
| Benchmark | GLM-5 | Claude Opus 4.5 | Gemini 3.0 Pro | GPT-5.2 | DeepSeek-V3.2 | Kimi K2.5 |
|---|---|---|---|---|---|---|
| SWE-bench Verified | 77.8% | 80.9% | 76.2% | 80.0% | 73.1% | 76.8% |
| SWE-bench Multilingual | 73.3% | 77.5% | 65.0% | 72.0% | 70.2% | 73.0% |
Takeaway:
GLM-5 is within striking distance of the best closed models in real-world software tasks.
For an open-weight model, that margin is small.
Agent & Tool Use
| Benchmark | GLM-5 | Claude Opus 4.5 | Gemini 3.0 Pro | GPT-5.2 | DeepSeek-V3.2 | Kimi K2.5 |
|---|---|---|---|---|---|---|
| BrowseComp | 62.0 | 37.0 | 37.8 | — | 51.4 | 60.6 |
| BrowseComp (Context Mgmt) | 75.9 | 67.8 | 59.2 | 65.8 | 67.6 | 74.9 |
| Vending Bench 2 ($) | $4,432 | $4,967 | $5,478 | $3,591 | $1,034 | $1,198 |
What stands out:
GLM-5 performs extremely well in agent-style and multi-step tasks especially compared to several closed systems.
It’s not the absolute top performer in Vending Bench 2, but it’s clearly operating in the same performance band.
The Bigger Picture
GLM-5 isn’t sweeping every single category. But It’s consistently competitive across reasoning, coding, and agent benchmarks at the same time.
That’s rare & when you factor in:
- Open weights
- MIT license
- Lower cost
It stops being “good for open.”
It becomes a serious alternative.
Wrapping Up
It’s not always about benchmark charts. Numbers matter but they’re only part of the story.
When you look closely at what GLM-5 can actually do, you start to see how far open-weight models have come.
And at this pace?
It’s not unrealistic to imagine a future where open models don’t just compete with closed ones, they surpass them.
That’s the bigger shift happening here.
