📚 New to this topic? Read our full guide: What Is Artificial Intelligence?.
Snowflake CEO Says China’s GLM-5.2 Nearly Matches Opus 4.7 at a Fraction of the Cost
The boss of a big data company just ran a test. The results are bad news for AI labs that charge a lot of money. Sridhar Ramaswamy is the CEO (the top leader) of Snowflake. He tried out a cheap Chinese AI model called GLM-5.2. It did coding work almost as well as a costly model called Claude Opus 4.7. But it cost far less money. He shared the test on June 24, 2026. GLM-5.2 is made by Zhipu, a Chinese AI company. Zhipu is also called Z.ai.
Why does this matter? Price is now a big reason people pick one AI over another. If a cheaper AI gives almost the same result, companies will switch to it. Ramaswamy’s test gives real, honest numbers from inside a big company. That is rare.
The test: 103 coding tasks, run three times
Ramaswamy set up a fair test. He gave both AI models the same 103 coding tasks. Each task was about writing code for two databases. (A database is a system that stores and organizes data.) The two were called DuckDB and Snowflake. To be fair, he ran every task three times.
The main result was very close. When each model got three tries, GLM-5.2 solved 66% of the tasks. Opus 4.7 solved 67%. That is almost a tie. But on the very first try, Opus did better. It got 53.7% right. GLM got 47.6% right. This kind of test is called a benchmark. (A benchmark is a fixed set of tasks used to compare AI models in a fair way.)
The real story is the price
The biggest difference is cost. AI models charge money by “tokens.” (Tokens are small pieces of text, roughly parts of words. You pay for each million tokens. You pay for what you send in and for what the model writes back.) GLM-5.2 charges far less per million tokens.
Benchmarks & specs: GLM-5.2 vs Opus 4.7
| Metric (as reported) | GLM-5.2 (Zhipu) | Claude Opus 4.7 | GPT-5.5 |
|---|---|---|---|
| Accuracy with 3 tries | 66% | 67% | — |
| First-attempt accuracy | 47.6% | 53.7% | — |
| Input price (per 1M tokens) | $1.40 | $5.00 | $5.00 |
| Output price (per 1M tokens) | $4.40 | $25.00 | $30.00 |
| Avg. runs per task | 99 | 80 | — |
| Total tokens used in test | 860 million | 439 million | — |
Where GLM-5.2 falls short
Cheaper does not mean perfect. Ramaswamy liked that GLM checked its code well across both databases. But he also saw real weak spots. Sometimes the model gave up on a task too soon. It also used too many tool calls. (A tool call is when the AI uses an outside tool, like running some code, to help solve a task.)
One example stood out. On a single task, GLM made 411 tool calls over 24 minutes. And it still failed. Opus solved the same task with just 49 calls in 9 minutes. So GLM can be slower and messier. This is true even when the final score is close.
Why “open-weight” Chinese models are a big deal
GLM-5.2 is one of many strong, low-cost models now coming out of China. A lot of them are “open-weight.” (Open-weight means the company shares the trained model. Other people can then run it on their own, often for less money.) This is the opposite of “closed” US models like Opus. With a closed model, you can only use it through the maker’s own service.
This change is shaking up the whole market. For years, the best AI came from just a few US labs. And it was expensive. Now Chinese rivals are catching up fast on quality. And they charge far less. This forces every AI company to defend its prices. It also gives buyers a real choice for the first time.
Ramaswamy’s test is useful because it is honest and detailed. He did not just trust a fancy marketing chart. He ran the same hard tasks on both models. Then he measured how accurate they were, how fast they were, how many tokens they used, and what they cost. This is the kind of real-world test that businesses should copy before they pick a model.
FAQ
Who makes GLM-5.2?
It is made by Zhipu, a Chinese AI company. It is also known as Z.ai.
Is GLM-5.2 better than Claude Opus 4.7?
Not quite. In this coding test, Opus 4.7 was a little more accurate. It was also far more efficient (it did the work using less time and fewer steps). But GLM-5.2 came very close, and it cost much less.
How much cheaper is GLM-5.2?
GLM-5.2 charges $1.40 to send text in and $4.40 to get text out, per million tokens. Opus 4.7 charges $5.00 to send in and $25.00 to get out. So GLM is several times cheaper per token.
Why it matters (especially for India and founders)
For Indian startups, cost is often what decides the choice. A model that is “good enough” and 5 times cheaper can change everything for a small team. It means AI features that once looked too costly may now be cheap enough to use. This is part of a bigger trend. Cheaper rivals are crowding into a market once ruled by US labs. At the same time, American firms are fighting to control how their top AI models spread. So founders should test cheaper models on their own real tasks before they spend money.
The main point: the gap between costly Western models and cheap Chinese open models is closing fast. The accuracy is close. The price is not. If you are building with AI, the smart move is simple. Test more than one model. Then pay only for the quality you really need.
Source: The Decoder.
Related coverage
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
