tl;dr

• Can GLM 4.7 beat Opus 4.5?
• Is MiniMax M2.1 worth trying?
• Qwen Image Edit gets an update

Releases

GLM 4.7

Chinese AI lab Z.ai has released an updated version of their open source coding model GLM 4.6.

The GLM models (and more specifically, the GLM coding plan) are known for their decent performance and incredibly low cost and great rate limits. The model can be dropped in directly to Claude Code as well as a wide variety of other agentic coding platforms for as low as $3 a month.

The 4.6 model was around the same level as Sonnet 4, and now the 4.7 version appears to be Sonnet 4.5 or maybe even a little better.

In my usage it has much better instruction following and can complete much larger tasks than its predecessor was able to.

It also is noticeably better at frontend design, being behind only Gemini 3 in Design Arena.

This would be much more exciting if it weren’t for Opus 4.5. GLM 4.7, or any other model, cannot hold a candle to Opus right now, making this a battle for second place. If you only need a second place model or have not been spoiled with Opus 4.5, then I would highly recommend checking this out, as it is the same quality as Sonnet 4.5 for

Model	$ per million (input)	$ per million (output)	Tokens per second
Claude Sonnet 4.5	$3	$15	57
Claude Opus 4.5	$5	$25	64
GPT 5.2	$1.75	$14	38
Gemini 3 Pro Preview	$2	$12	80
GLM 4.7	$0.60	$2.20	280

Data from Openrouter

MiniMax 2.1

Another big release from China this week is the updated version of MiniMax 2. Minimax is the same lab that makes the Hailuo video generation models, which are always near the top of the video generation leaderboards.

The Minimax 2 model is the Gemini Flash of the Chinese models, being smaller and cheaper than most other models while still being very strong. It is a 230 billion parameter mixture of experts model, with 10 billion active parameters. For reference, DeepSeek is 685 billion params with 37 billion active, and GLM 4.7 is 358 billion total with 32 billion active. This means that with the same hardware, MiniMax will be ~3 times faster than the other two models.

Model	$ per million (input)	$ per million (output)	Tokens per second
GLM 4.7	$0.60	$2.20	280
MiniMax M2.1	$0.30	$1.20	66
DeepSeek 3.2	$0.28	$0.42	29

It is ridiculous how cheap DeepSeek is given its size

The model is being pushed as a coding model, and is decent from what I have seen, being around the GLM 4.6 level, which would be at the top for Chinese models, if it weren’t for GLM 4.7 being released within a week as well.

It does have a chance of being run on home servers, as it should fit on a computer with 128GB of memory with 4 bit weights, which none of the other frontier Chinese models can do.

Overall it seems to be a decent model, fairly cheap, and fast, but nothing special that we do not already have.

Quick Hits

Qwen Image Edit Update

Qwen Image Edit is the most popular open source image editing model right now.

When it was released, it was in the top 3 for image editing models, but the competition in the space has been fierce and it has dropped outside of the top 10.

However, because it is open source, the community has been making fine tuned versions that have increased its capabilities for narrow tasks.

What the Qwen team has done for this release is to add in many of the popular community finetunes and then also add some more of their own training on top, resulting in a stronger model.

Official benchmark scores are not out yet, but I suspect we will be seeing Qwen Image Edit back at the top again.

Finish

Happy holidays! I hope you enjoyed the news this week. If you want to get the news every week, be sure to join our mailing list below.

From Jocelin Carmes on Twitter